CN101751439A - Image retrieval method based on hierarchical clustering - Google Patents
Image retrieval method based on hierarchical clustering Download PDFInfo
- Publication number
- CN101751439A CN101751439A CN200810240361A CN200810240361A CN101751439A CN 101751439 A CN101751439 A CN 101751439A CN 200810240361 A CN200810240361 A CN 200810240361A CN 200810240361 A CN200810240361 A CN 200810240361A CN 101751439 A CN101751439 A CN 101751439A
- Authority
- CN
- China
- Prior art keywords
- phrase
- image
- cluster
- key word
- 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 27
- 230000000007 visual effect Effects 0.000 claims abstract description 18
- 239000000284 extract Substances 0.000 claims description 11
- 239000000463 material Substances 0.000 claims description 6
- 238000010586 diagram Methods 0.000 description 7
- 230000008569 process Effects 0.000 description 7
- 230000003247 decreasing effect Effects 0.000 description 5
- 230000000694 effects Effects 0.000 description 5
- 241000282326 Felis catus Species 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 241000282376 Panthera tigris Species 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 235000013399 edible fruits Nutrition 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 241000581835 Monodora junodii Species 0.000 description 1
- 241001323321 Pluto Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 230000002354 daily effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000012797 qualification Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000005303 weighing Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
Abstract
The invention relates to an image retrieval method based on hierarchical clustering, proposes a system schema based on hierarchical clustering network image retrieval on the basis of the traditional image search engine through keyword semantic analysis and image visual clustering, and mainly includes that: (1) the image retrieval results are clustered in semantic level; (2) the image retrieval results are clustered in visual characteristic level; and (3) the hierarchical clustering navigation display is realized in a rapid and efficient way. Compared with the traditional image search engine, the invention adds the clustering idea to the image retrieval. The keywords are semantically analyzed to form different subjects; meanwhile, the images are also analyzed in visual characteristic level to cluster similar images in a class, and finally the retrieval results are displayed for the user in classification mode through a concise and clear interface, so as to help users quickly and efficiently find out the required target image from the retrieval results with unsorted subjects.
Description
Technical field
The invention belongs to technical field of image processing, relate to a kind of image search method based on hierarchical clustering.
Background technology
Along with popularizing and the development of digital media technology of internet, applications, the most important carrier that image transmits as information has been deep into the every aspect of people's daily life.All can there be every day number to pour in the internet, how these images be carried out effective organization and management, allow the user find needed information to become a great problem of needing solution at present badly fast and accurately with the image of GB even TB.
Traditional image search engine such as Google, Yahoo etc. mostly according to the relevant textual information of network image and the degree of correlation of searching keyword, through ordering, present to the user with result for retrieval.Yet, because the polysemy of key word, comprise a plurality of themes usually in the result for retrieval and they are aliasing in together alternately.As Fig. 1 is preceding 16 results of key word " apple " in the Google image search engine, with " apple " " relevant speech has " apple computer ", " apple fruit ", " applelogo ", " apple iPod ", " apple phone " or the like.Finding the content that oneself needs from the results list that these themes mix, will be a job of wasting time and energy.
Summary of the invention
The problem to be solved in the present invention: be to find a kind ofly can be automatically carry out the method that theme is sorted out, allow the user conveniently find the information of own needs, for this reason, the purpose of this invention is to provide a kind of image search method based on hierarchical clustering to result for retrieval.
For reaching described purpose, the image search method based on hierarchical clustering provided by the invention comprises that step is as follows:
Step 1: use the key word text search, the document relevant with key word that obtains analyzed, extract the phrase relevant,, obtain semantic cluster to these phrases cluster from the semantic level with key word;
Step 2:, obtain the cluster of picture material to image searching result cluster on the visual signature aspect;
Step 3: on the basis of search engine retrieving result display interface, add a hierarchical clustering navigation bar, the hierarchical clustering navigation that is used for convenient and efficient shows.
Embodiments of the invention, cluster is extracted the relevant phrases feature of image searching result from the described semantic level, for any one given key word, obtains the document relevant with key word by text search engine earlier; From these documents, extract the phrase relevant then with key word, note simultaneously phrase appearance in the document frequency, comprise the document ratio of phrase, the length information of phrase; Use these information of recurrence learning model generalization at last, be converted to scoring these phrase correlativitys, like this, a preceding n phrase be to look for the maximally related correlativity phrase of key word.
Embodiments of the invention, cluster is extracted the similarity degree between the phrase from the described semantic level, use and carry out cluster based on the method for k-line, to the similarity degree NGD between the phrase (x, y) weigh with following formula:
NGD(x，y)＝(max{logf(x)，logf(y)}-logf(x，y))/(logN-min{logf(x)，logf(y)})
Wherein f (x) and f (y) represent respectively to retrieve the number of pages of returning as a result with phrase x or y in the Google search engine separately, f (x, y) expression is put into the number of pages as a result that retrieval is returned in the Google search engine together with phrase x and y, two phrases are similar more, and to unite the probability of appearance big more, similarity degree NGD (x, y) more little, the method cluster that re-uses like this based on K-line just can gather one group to the very big phrase of correlativity, form a theme, just can be shown to the user, allow the information that the user conveniently finds oneself to be needed according to the classifying importance of theme.
Embodiments of the invention, cluster on described visual signature aspect is to carry out on the basis of Semantic Clustering, at first retrieval obtains the image relevant with each phrase, extract their visual signature then, and calculate correlativity between each image, utilize these information to carry out the cluster of picture material at last.
Embodiments of the invention, the described navigation of hierarchical clustering efficiently shows, on the basis of traditional search engines result for retrieval display interface, add a hierarchical clustering navigation bar, this navigation bar will be relevant with key word image searching result according to the importance of theme, according to the correlativity on the vision aspect, be shown to the user disaggregatedly.Such surface structure allows the user be easy to just focus on the visual effect of own topics of interest and oneself needs, helps the user to find own needed target image fast and efficiently from the result for retrieval of theme aliasing.
Beneficial effect of the present invention: the present invention is different with the traditional images search engine, is that cluster thought is dissolved into image retrieval.Key word is carried out semanteme resolve, form different themes; Simultaneously image is analyzed from visual signature, close image is gathered in a class, interface by simple and clear is shown to the user with result for retrieval categorizedly at last, thereby helps the user to find own needed target image quickly and efficiently from the result for retrieval of theme aliasing.
Description of drawings
Preceding 16 results of Fig. 1 prior art key word " apple " in the Google of Google image search engine;
Fig. 2 is an integral frame process flow diagram of the present invention;
Fig. 3 extracts the process flow diagram of the relevant phrases of key word for the present invention;
Fig. 4 system user interactive interface;
The vision cluster of Fig. 5 " macbook pro " shows;
Fig. 6 the present invention and IGroup system of Microsoft and the Google of system of Google contrast effect;
The effort that Fig. 7 user search informational needs is paid relatively;
Fig. 8 retrieval effectiveness comparison diagram;
Embodiment
Describe each related detailed problem in the technical solution of the present invention in detail below in conjunction with accompanying drawing.Be to be noted that described embodiment only is intended to be convenient to the understanding of the present invention, and it is not played any qualification effect.
The present invention has proposed a kind of framework of hierarchical clustering from semantically resolving key word, analyzing result for retrieval from visual signature, automatically result for retrieval is carried out cluster, and returns to the user according to theme, allows the information that the user conveniently finds oneself to be needed.As Fig. 2 three parts that system's (abbreviation native system) integral frame process flow diagram of realizing with the inventive method comprises are shown: (1) from the semantic hierarchies to the result for retrieval cluster of image; (2) on the visual signature level to the image searching result cluster; (3) navigation of the hierarchical clustering of user's convenient and efficient is mutual shows.Native system is all worked all to develop on a computing machine and is finished, system adopt the JAVA language realized web crawlers climb get network image make up database, realized the row's of falling mechanism set up index, realized characteristics of image extraction, realized cluster on the semantic hierarchies and the cluster on the visual signature level, adopt JSP and html language to make up that webpage is used for the user interactions input and last result shows.
Cluster comprises three steps on the semantic hierarchies:
The first, resolve key word from semantic hierarchies, extract the phrase relevant with key word.See also Fig. 3 and extract the process flow diagram of the relevant phrases of key word for the present invention, for any one given key word, the document that obtains being correlated with of text search engine such as the Google of Google, the Yahoo of Yahoo, the Baidu of Baidu etc. by present comparative maturity earlier; From these documents, extract the phrase relevant then, note length or the like the information of the frequency of phrase appearance in the document, the document ratio that comprises phrase, phrase simultaneously with key word; Use these information of recurrence learning model generalization at last, be converted to scoring these phrase correlativitys.Preceding n the phrase that obtains like this be exactly we to look for the maximally related phrase of key word.
The second, after another feature of cluster then is to obtain relevant phrases on the semantic level, uses and these phrases are carried out cluster based on the method for k-line, each class is just represented a theme here.Semantic Clustering process based on K-line is as follows:
| Calculating similarity matrix: A=exp (NGD (and x, y) 2/2δ 2) wherein: δ 2＝[mean(NGD(x，y))] 2/ 2 2. calculate Laplce's matrix: L=S -1/2AS -1/2Wherein: S ii＝∑ jA ij S when i ≠ j ij=0 3. calculates preceding K proper vector m 1...m k4. all points are divided into according to the distance of they and proper vector and come 5. in k the class, create matrix M for any class j j＝[y i] i∈pj, use M jM j TEigenvalue of maximum characteristic of correspondence vector upgrade m jWherein: p jThe institute that represents the j class has a little 6. to repeat 4,5 and go on foot up to m jNo longer change |
Wherein, NGD (Normalized Google Distance) is the correlativity that is used for weighing between a pair of phrase x and the y, the NGD of the similarity degree between the phrase (x, y) formula is weighed:
N represents the number of the total document of Google search engine, f (x) and f (y) represent respectively to retrieve the number of pages of returning as a result with phrase x or y in the Google search engine separately, f (x, y) expression is put into the number of pages as a result that retrieval is returned in the Google search engine together with phrase x and y, according to following formula obviously, two phrases are similar more, the probability that they unite appearance is big more, NGD (x, y) more little, the method cluster that re-uses like this based on K-line just can gather one group to the very big phrase of correlativity, forms a theme.
The 3rd, result for retrieval is divided into different themes after, successively decreasing successively according to the importance of theme is shown to the user.The importance SImportance of theme (m) is defined as follows:
N wherein
mBe the number of phrase in the m class, ResultSize (i) is i the picture number that phrase obtains.
From top process as can be seen, cluster not only is divided into result for retrieval different themes on the semantic hierarchies, can also be shown to the user successively by the importance difference of theme.Therefore, it can make the user focus on the target topic that will look for quickly and efficiently itself.
Yet, only carry out cluster not enough, because for wherein some themes, the image list that obtains also will be rich and varied from vision content from semantic hierarchies.Such as: the image list that " apple fruit " obtains may comprise " red apple " and " green apple ", and this just makes the cluster on the picture material level necessitate.
Cluster on the picture material level is primarily aimed at the image collection relevant with each phrase, extracts their visual signature such as feature such as color, texture, shape respectively, adopts improved K-means method to carry out image clustering based on vision content then.Concrete cluster process is as follows:
| 1. select K object as initial center arbitrarily from N data object, simultaneously, choose the maximum number S of object in the class.2. for other last objects,, distribute to the cluster the most similar to it according to the similarity at they and these class centers.3. for the new class that is obtained, choose S the object nearest and upgrade the class center from current class center.4. constantly repeating 2 and 3 no longer changes and ends up to each cluster. |
After this cluster was finished, the user was presented in the representative that is chosen as this vision class with nearest that image in class center, other image according to the arrangement of successively decreasing successively of the distance at class center.Each vision class is according to the arrangement of successively decreasing of its importance, and its importance may be defined as:
Vlmportance(n)＝ClusterSize(n)/σ
n
Wherein, the number of the image that is comprised in n vision class of ClusterSize (n) expression, σ
nThe standard deviation of representing n vision class.
In sum, the cluster of image vision content is on the basis of Semantic Clustering, and the relevant image of each theme is organized and analyzed, and it is gathered into several vision classes, chooses simultaneously and the representative of the nearest image in center as this vision class, presents to the user.Like this, can be so that the user can focus on certain vision class easily, and find the target image that oneself needs rapidly.
In order better to show the effect of this hierarchical clustering, also designed User Interface as shown in Figure 4 here.It is made up of three parts: user's input window (QView), hierarchical clustering navigator window (HCView), result for retrieval display window (RView).On the basis of traditional search engines result for retrieval display interface, added a hierarchical clustering navigation bar.This navigation bar not only will be relevant with key word image searching result according to the importance of theme, according to the correlativity on the vision aspect, be shown to the user disaggregatedly, surface structure that the more important thing is simple and clear allows the user be easy to just focus on the visual effect of own topics of interest and oneself needs, can help the user to find own needed target image quickly and efficiently from the result for retrieval of theme aliasing.
User's input window is positioned at the top of whole framework, and the user needs the information of any aspect can use the automatic retrieval of edit box input key word to obtain.Remove this, native system is also supported query composition, such as " dog AND cat ", " dog OR cat ", " dog NOT cat " or the like.The hierarchical clustering navigator window is positioned at the left side of framework.Phrase with similar semanteme shows in a red square frame that all promptly the class in the Semantic Clustering is represented a theme.Each semantic category is according to the arrangement of successively decreasing of their importance.Simultaneously, to each phrase, we carry out the vision cluster to it, and the representative of each vision class are presented at the below of this phrase with the situation of thumbnail.The vision class is also according to the arrangement of successively decreasing of their importance.When the phrase relevant with key word is clicked, will assist at the RView window to show all vision classes under this phrase, as the vision cluster demonstration of Fig. 5 " macbook pro ".After if the representative image under the key phrase is clicked, the different all images that show under this vision class of its degree of correlation will pressed under the RView window.
Implementation result
In order better to assess native system, we have designed three experiments, from the user performance of this system of comprehensive evaluation is tested in the satisfaction of total system, effort and three aspects of retrieval effectiveness that user search need be paid to information needed to this system respectively.This is tested all data and all swashes from Google and Flickr and get, the text message of using in the test comprises title, URL, ALT mark, link anchor text and the image text on every side of image, and the visual signature of using comprises the color histograms of 144 dimensions and PWT (the Polynomial Wavelet Tree) features of 24 dimensions.For the satisfaction of assesses user to this total system, we have invited a plurality of users to retrieve with any speech on the Google of system of Google, IGroup system of Microsoft and native system HICluster respectively, and feed back their satisfaction to these three systems.Here use three yardsticks to come the satisfaction of comparison user to different system: "=" expression two system effects are identical; The result of the previous system of "＞" expression is better than back one; "＞＞" represent that the former is much better than the latter.In order better to assess satisfaction, we quantize these three yardsticks.Least Man Yi system is assigned 1, and simultaneously: if X=Y, then both give identical value; If X＞Y, then the value of X is to add 1 on the basis of Y, and if X＞＞Y, then the value of X adds 2 on the basis of Y.Fig. 6 has shown the effect of native system, as can be seen from the figure, the average of three systems, the user is better than IGroup system of Microsoft and the Google of system of Google to the satisfaction of native system.
For assesses user retrieves the effort that information needed need be paid, we please different users do experiment with a plurality of speech respectively on the Google of system of Google, IGroup system of Microsoft and native system HICluster in a period of time, and feed back them and retrieve the effort that demand information need be paid.Here being used for measuring the parameter of working hard comprises: the number of times (N of the query word of input
q), the picture number of links (N that clicks
Plc), the IGroup of the Microsoft system class number (N that clicks
Cnc), the number (N of the native system HICluster semantic category clicked
Scc), the number (N of the native system HICluster vision class clicked
Vcc).The criterion of working hard is as follows:
SE
Google＝N
q+N
plc
SE
IGroup＝N
q+N
plc+N
cnc
SE
HICluster＝N
q+N
plc+N
scc+N
vcc
Fig. 7 has provided the experiment comparison diagram of the effort that the user search informational needs pays, and native system HICluster is primarily aimed at English retrieval, the apple that occurs among the figure, tiger, pluto at present ... Deng speech is the key word of target search.As can be seen from the figure, for different speech, native system HICluster can allow the user find required information more easily with respect to IGroup system of Microsoft and the Google of system of Google.Especially, with respect to key word " dog " and " cat ", the advantage of native system on " apple " and polysemants such as " tiger " is more obvious, and this explanation native system HICluster has reduced the difficulty of polysemant retrieval effectively.In order to assess retrieval effectiveness, here the picture number that retrieves, the picture number that the vision class covers, the picture number that semantic category covers are analyzed.The result is the retrieval effectiveness comparison diagram as shown in Figure 8, the total number of images that covers from the detected associated picture sum of the HICluster of native system shown in Fig. 8, IGroup system of Microsoft vision class, the total number of images that the Google of system of Google semantic category covers: at first, native system HICluster (M=17.2, SD=2.1) total number of images of giving for change is than (the M=12.3 of IGroup system of Microsoft, SD=1.8) and the Google (M=10.6 of system of Google, SD=2.3) many, this has shown that the hierarchical clustering structure of native system HICluster makes the user find relevant image more easily.Secondly, native system HICluster (M=5.2, SD=1.9) (the M=4.5 of IGroup system of vision analogy Microsoft, SD=2.1) and the Google (M=3.2 of system of Google, SD=2.2) covered more images, this has shown that native system HICluster can offer how different, relevant with the target image of user.At last, the image of three system semantics class coverings is close, and this also is reasonable, because the user is before submitting search key to, the target that retrieve has been had understanding clearly.
The above; only be the embodiment among the present invention; but protection scope of the present invention is not limited thereto; anyly be familiar with the people of this technology in the disclosed technical scope of the present invention; can understand conversion or the replacement expected; all should be encompassed in of the present invention comprising within the scope, therefore, protection scope of the present invention should be as the criterion with the protection domain of claims.
Claims (5)
1. the image search method based on hierarchical clustering is characterized in that, comprises that step is as follows:
Step 1: use the key word text search, the document relevant with key word that obtains analyzed, extract the phrase relevant,, obtain semantic cluster to these phrases cluster from the semantic level with key word;
Step 2:, obtain the cluster of picture material to image searching result cluster on the visual signature aspect;
Step 3: on the basis of search engine retrieving result display interface, add a hierarchical clustering navigation bar, the hierarchical clustering navigation that is used for convenient and efficient shows.
2. image search method according to claim 1, it is characterized in that, cluster is extracted the relevant phrases feature of image searching result from the described semantic level, for any one given key word, obtains the document relevant with key word by text search engine earlier; From these documents, extract the phrase relevant then with key word, note simultaneously phrase appearance in the document frequency, comprise the document ratio of phrase, the length information of phrase; Use these information of recurrence learning model generalization at last, be converted to scoring these phrase correlativitys, like this, a preceding n phrase be to look for the maximally related correlativity phrase of key word.
3. image search method according to claim 1, it is characterized in that cluster is extracted the similarity degree between the phrase from the described semantic level, use and carry out cluster based on the method for k-line, to the similarity degree NGD between the phrase (x, y) weigh with following formula:
NGD(x，y)＝(max{logf(x)，logf(y)}-logf(x，y))/(logN-min{logf(x)，logf(y)})
Wherein f (x) and f (y) represent respectively to retrieve the number of pages of returning as a result with phrase x or y in the Google search engine separately, f (x, y) expression is put into the number of pages as a result that retrieval is returned in the Google search engine together with phrase x and y, two phrases are similar more, and to unite the probability of appearance big more, similarity degree NGD (x, y) more little, the method cluster that re-uses like this based on K-line just can gather one group to the very big phrase of correlativity, form a theme, just can be shown to the user, allow the information that the user conveniently finds oneself to be needed according to the classifying importance of theme.
4. image search method according to claim 1, it is characterized in that, cluster on described visual signature aspect is to carry out on the basis of Semantic Clustering, at first retrieval obtains the image relevant with each phrase, extract their visual signature then, and calculate correlativity between each image, utilize these information to carry out the cluster of picture material at last.
5. described image search method according to claim 1, it is characterized in that, the described navigation of hierarchical clustering efficiently shows, on the basis of traditional search engines result for retrieval display interface, add a hierarchical clustering navigation bar, this navigation bar will be relevant with key word image searching result according to the importance of theme, according to the correlativity on the vision aspect, be shown to the user disaggregatedly.Such surface structure allows the user be easy to just focus on the visual effect of own topics of interest and oneself needs, helps the user to find own needed target image fast and efficiently from the result for retrieval of theme aliasing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN200810240361A CN101751439A (en) | 2008-12-17 | 2008-12-17 | Image retrieval method based on hierarchical clustering |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN200810240361A CN101751439A (en) | 2008-12-17 | 2008-12-17 | Image retrieval method based on hierarchical clustering |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN101751439A true CN101751439A (en) | 2010-06-23 |
Family
ID=42478430
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN200810240361A Pending CN101751439A (en) | 2008-12-17 | 2008-12-17 | Image retrieval method based on hierarchical clustering |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101751439A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102375824A (en) * | 2010-08-12 | 2012-03-14 | 富士通株式会社 | Device and method for acquiring multilingual texts with mutually corresponding contents |
| CN102521233A (en) * | 2010-11-02 | 2012-06-27 | 微软公司 | Adaptive image retrieval database |
| US8429163B1 (en) | 2012-01-25 | 2013-04-23 | Hewlett-Packard Development Company, L.P. | Content similarity pyramid |
| CN103207879A (en) * | 2012-01-17 | 2013-07-17 | 阿里巴巴集团控股有限公司 | Method and equipment for generating image index |
| CN104317867A (en) * | 2014-10-17 | 2015-01-28 | 上海交通大学 | System for carrying out entity clustering on web pictures returned by search engine |
| CN104765768A (en) * | 2015-03-09 | 2015-07-08 | 深圳云天励飞技术有限公司 | Mass face database rapid and accurate retrieval method |
| CN107577682A (en) * | 2016-07-05 | 2018-01-12 | 上海交通大学 | Users' Interests Mining and user based on social picture recommend method and system |
| CN109101542A (en) * | 2018-07-02 | 2018-12-28 | 深圳市商汤科技有限公司 | Image recognition result output method and device, electronic equipment and storage medium |
| CN110597719A (en) * | 2019-09-05 | 2019-12-20 | 腾讯科技（深圳）有限公司 | Image clustering method, device and medium for adaptation test |
| CN113191411A (en) * | 2021-04-22 | 2021-07-30 | 杭州卓智力创信息技术有限公司 | Electronic sound image file management method based on photo group |
| WO2021208633A1 (en) * | 2020-04-16 | 2021-10-21 | 腾讯科技（深圳）有限公司 | Method and device for determining item name, computer apparatus, and storage medium |
-
2008
- 2008-12-17 CN CN200810240361A patent/CN101751439A/en active Pending
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102375824B (en) * | 2010-08-12 | 2014-10-22 | 富士通株式会社 | Device and method for acquiring multilingual texts with mutually corresponding contents |
| CN102375824A (en) * | 2010-08-12 | 2012-03-14 | 富士通株式会社 | Device and method for acquiring multilingual texts with mutually corresponding contents |
| US9317533B2 (en) | 2010-11-02 | 2016-04-19 | Microsoft Technology Licensing, Inc. | Adaptive image retrieval database |
| CN102521233A (en) * | 2010-11-02 | 2012-06-27 | 微软公司 | Adaptive image retrieval database |
| US9330341B2 (en) | 2012-01-17 | 2016-05-03 | Alibaba Group Holding Limited | Image index generation based on similarities of image features |
| CN103207879A (en) * | 2012-01-17 | 2013-07-17 | 阿里巴巴集团控股有限公司 | Method and equipment for generating image index |
| TWI552007B (en) * | 2012-01-17 | 2016-10-01 | Alibaba Group Holding Ltd | Method and apparatus for generating image index |
| CN103207879B (en) * | 2012-01-17 | 2016-03-30 | 阿里巴巴集团控股有限公司 | The generation method and apparatus of image index |
| US8429163B1 (en) | 2012-01-25 | 2013-04-23 | Hewlett-Packard Development Company, L.P. | Content similarity pyramid |
| CN104317867B (en) * | 2014-10-17 | 2018-02-09 | 上海交通大学 | The system that entity cluster is carried out to the Web page picture that search engine returns |
| CN104317867A (en) * | 2014-10-17 | 2015-01-28 | 上海交通大学 | System for carrying out entity clustering on web pictures returned by search engine |
| CN104765768A (en) * | 2015-03-09 | 2015-07-08 | 深圳云天励飞技术有限公司 | Mass face database rapid and accurate retrieval method |
| CN107577682A (en) * | 2016-07-05 | 2018-01-12 | 上海交通大学 | Users' Interests Mining and user based on social picture recommend method and system |
| CN107577682B (en) * | 2016-07-05 | 2021-06-29 | 上海交通大学 | Social picture-based user interest mining and user recommending method and system |
| CN109101542A (en) * | 2018-07-02 | 2018-12-28 | 深圳市商汤科技有限公司 | Image recognition result output method and device, electronic equipment and storage medium |
| CN110597719A (en) * | 2019-09-05 | 2019-12-20 | 腾讯科技（深圳）有限公司 | Image clustering method, device and medium for adaptation test |
| WO2021208633A1 (en) * | 2020-04-16 | 2021-10-21 | 腾讯科技（深圳）有限公司 | Method and device for determining item name, computer apparatus, and storage medium |
| CN113191411A (en) * | 2021-04-22 | 2021-07-30 | 杭州卓智力创信息技术有限公司 | Electronic sound image file management method based on photo group |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101751439A (en) | Image retrieval method based on hierarchical clustering | |
| CN110008311B (en) | Product information safety risk monitoring method based on semantic analysis | |
| CN109992645B (en) | Data management system and method based on text data | |
| JP6416150B2 (en) | Search method, search system, and computer program | |
| CN102253982B (en) | Query suggestion method based on query semantics and click-through data | |
| CN105045875B (en) | Personalized search and device | |
| CN110334178B (en) | Data retrieval method, device, equipment and readable storage medium | |
| CN109271477A (en) | A kind of method and system by internet building taxonomy library | |
| CN101609450A (en) | Web page classification method based on training set | |
| CN106202514A (en) | Accident based on Agent is across the search method of media information and system | |
| CN104317834A (en) | Cross-media sorting method based on deep neural network | |
| CN103544255A (en) | Text semantic relativity based network public opinion information analysis method | |
| CN102254039A (en) | Searching engine-based network searching method | |
| CN107885793A (en) | A kind of hot microblog topic analyzing and predicting method and system | |
| CN102902806A (en) | Method and system for performing inquiry expansion by using search engine | |
| KR101355945B1 (en) | On line context aware advertising apparatus and method | |
| Bin et al. | Web mining research | |
| CN115796181A (en) | Text relation extraction method for chemical field | |
| KR100954842B1 (en) | Method and System of classifying web page using category tag information and Recording medium using by the same | |
| CN115905489B (en) | Method for providing bidding information search service | |
| CN107908749B (en) | Character retrieval system and method based on search engine | |
| CN111966899A (en) | Search ranking method, system and computer readable storage medium | |
| CN103823847A (en) | Keyword extension method and device | |
| CN114238735B (en) | Intelligent internet data acquisition method | |
| US9305103B2 (en) | Method or system for semantic categorization |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C12 | Rejection of a patent application after its publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20100623 |