CN103514254A - Image set ordering method for mining hidden operation behavior - Google Patents
Image set ordering method for mining hidden operation behavior Download PDFInfo
- Publication number
- CN103514254A CN103514254A CN201310279399.7A CN201310279399A CN103514254A CN 103514254 A CN103514254 A CN 103514254A CN 201310279399 A CN201310279399 A CN 201310279399A CN 103514254 A CN103514254 A CN 103514254A
- Authority
- CN
- China
- Prior art keywords
- picture
- pictures
- user
- image
- user preference
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/951—Indexing; Web crawling techniques
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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
- G06F16/5838—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using colour
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Databases & Information Systems (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Library & Information Science (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
Provided is an image set ordering method for mining a hidden operation behavior. According to the image set ordering method, the screening and ordering technology is improved, and users of different types can obtain search results suitable for individuals to the largest extent, so that user repeated search frequency and additional scoring operation frequency are reduced, search efficiency of the users is improved, and searching satisfaction of the users is improved.
Description
Technical field
The invention belongs to network picture search technique field, and in particular to a kind of image collection sort method of the recessive operation behavior of digging user, more particularly to a kind of method that the picture that screens and sort with feedback data is browsed according to user.
Background technology
In nearest decades, with the popularization of personal computer and network, one of important sources of people's acquisition information have been had changed into by web search message.Wherein, picture searching is an important component of web search.All the time, existing picture searching often filters out the result being not very consistent with demand to user.This is due to the following reasons what is caused：First, the overwhelming majority is that the algorithm of search results ranking is using essential informations such as pageviews as foundation, it is impossible to provide the user the Search Results of personalization.Because the demand of each user has significant difference, this general algorithm is doomed that Consumer's Experience can be reduced；Second, the association reponse system being recently formed, although the evaluation of reflection group of subscribers that can be relatively good to Search Results, but be not an individually individual in view of each user, also will it is variant the need for.3rd, the searching system that existing personalized feedback is adjusted, excessive can just provide accurate feedback result dependent on manual scoring of the user to picture, and extra operation needs are caused to user.It would therefore be desirable to a kind of picture screening and sort method, to provide the user the result closer to demands of individuals.
The content of the invention
The present invention provides a kind of image collection sort method of the recessive operation behavior of digging user, and the present invention passes through a kind of new picture set sort method.The present invention is by improving screening and ordering techniques, different types of user is accessed and be adapted to personal Search Results to greatest extent, the number of times of user's repeat search number of times and extra credits operation is reduced, user's search efficiency is improved, satisfaction of the user to search is improved.
To achieve the above object, the technical scheme is that：
A kind of image collection sort method of the recessive operation behavior of digging user, step is as follows：
Step 1：According to user when the content based on sample image carries out image retrieval the pictures that retrieve, with the picture browsing information of statistical tool counting user, including：Browsing time, download time and scaling, dragging number of times per pictures；
Step 2：After image statistics are obtained, for certain pictures, given a mark using below equation：
Wherein, V represents the marking result of single picture（Between 0.0 to 1.0）, i=1,2,3 represent the browsing time respectively, drag number of times, scaling number of times. 
Represent the judge weights of other three browse operation factors. Represent the normalized data of corresponding factor.H is definite value 1：When picture is not downloaded out-of-date, d=0 is made；When picture is downloaded out-of-date, d=1 is made, so ensures that the picture being downloaded scoring is 1 i.e. full marks, and the browsed picture do not downloaded can be according to
Standards of grading provide scoring；（Formula is not complete, it is impossible to show）
For
(i=1,2,3) we provide weights as follows：Collect browsing time, download time and scaling per pictures, drag number of times, form data matrix：
WhereinData after the use minimax value method normalization of the corresponding index of expression m pictures.Then, we seek data matrix F its covariance matrix C：
Then, we are by following equation, to calculate C characteristic value：
Wherein P is C characteristic vector,
It is C characteristic value.This equation is solved, we can obtain 3 solutions, i.e.,：
We are right
It is ranked up, obtains a new disaggregation：
According to
In sortord, the order of our three column vectors to F andCarry out corresponding
Sequence.According to many experiments data, we are corresponding to maximum
'sIt is assigned to 0.5, second largest for 0.3, minimum is assigned to 0.2.Then just can basis
Computing formula, come calculate its scoring；
According to appraisal result from high to low, the picture in pictures is divided into user preference picture subset and does not meet two subsets of user preference pictures（For a pictures, it is divided into according to its scoring V (I) using 0.4 for threshold value and meets preference and do not meet preference as standard）；
Step 3：It regard the subset of the not browsed similar image construction of user as test pictures collection in addition；
Step 4：Respectively to meeting user preference pictures, not meeting user preference pictures and three principal characters of test pictures collection extraction image：Boundary characteristic, textural characteristics and color characteristic, set up the eigenmatrix of each picture subset；
Step 5：After the eigenmatrix of picture subset is obtained, SVM training is carried out to the picture in test pictures collection first against three single features, obtain judging that the image in test pictures collection belongs to the grader for meeting user preference or not meeting user preference subset, and then obtain thinking the picture for meeting user preference；Three single features eigenmatrixes are merged into a total eigenmatrix again, SVM training is carried out to the picture in test pictures collection, obtains thinking the picture for meeting user preference；
Step 6：Every pictures in test pictures collection are all manually demarcated, determine whether every image actually meets user preference；
Step 7：Step will be passed through（5）Different characteristic combination the test pictures collection calibration result trained and step（6）The result manually demarcated is compared, and selectes the characteristic set for being best suitable for such pictures；
Step 8:Opened an account for each user for having record, image category and corresponding image feature information that its searching preferences, i.e. user searched for are recorded under the account.When user searches again for same retrieval input picture or similar image, according to existing user preference information, step is utilized（7）Optimal characteristics set and new information image searching result is trained, realize classification and reorder there is provided the picture close with its personal like to user, improve constantly precision by such iterative process.
The present invention also has the following advantages that：
The effectiveness and performance of the disaggregated model of the present invention are influenceed by two factors altogether：The quantity of the conspicuousness that different characteristic influences in image classification not of the same race, the number of times of user's same content search repeatedly, and the picture browsed in search procedure each time.We have carried out three groups of experiments to probe into the influence of these three factors altogether.In an experiment, we weigh these three factors using following two statistics to training result influence degree.
（1）Hit rate：The real amount of images for meeting user preference that screening system goes out accounts for the ratio of image total amount.
（2）Give rate for change：The real amount of images for meeting user preference that screening system goes out accounts for the ratio of the amount of images of actual coincidence user preference.
When the conspicuousness influenceed in research different characteristic in image classification not of the same race influences on model performance, we are tested seabeach, cloth, the earth, flower, ship, ripple pictures, obtained result is shown, seabeach, the maximally effective feature of flower pictures classification are color characteristics, cloth, the maximally effective feature of ripple pictures classification are textural characteristics, and ship and the maximally effective feature of earth pictures classification are boundary characteristics.This demonstrate that for different types of picture, with the most suitable feature of difference.It is contemplated that having arrived this point, the more accurate Search Results for being adapted to personal preference can be brought for user.
When studying user's searched same picture material repeatedly, we are experiment sample using six pictures, using user for same content different searching times as independent variable, probe into it for hit rate and the influence for giving rate for change.
Brief description of the drawings
Fig. 1 is simulation system explanation figure
Fig. 2 is meeting hobby, not meeting the precision statisticses of hobby for six test data sets
Fig. 3 is precision statisticses of six test data sets under different user's searching times
Fig. 4 is, to ripple pictures, to count different weight percentage（10%, 15%, 20%）User's browsing pictures under the conditions of, the precision statisticses under different user searching times
Embodiment
Below by specific embodiment, the present invention will be further described：
As shown in Figure 1, Figure 2, shown in Fig. 3 and Fig. 4, the image collection sort method of digging user recessiveness operation behavior, step is as follows：
1. the picture browsing information of counting user.
（1）Information is browsed because existing browser can not store our users of needs, is the simulation system that this present invention establishes an extraction user profile, as shown in Figure 1.Within the system, we have imported the pictures retrieved using other browsers, and browse information according to user behavior statistics is various, including：Browsing time, download time and user's scoring per pictures.Because user is when browsing Search Results, it will usually continuously browse result pictures, therefore it is considered that the number of visits of single picture is worth for picture screening and sequencing and without height.And browsing time and download time can significantly embody value of the pictures for user:User is more satisfied for picture, and the browsing time can be longer, it may have bigger possibility removes download pictures.User's scoring is most directly to reflect information of the user to image satisfaction.After image statistics are obtained, after image statistics are obtained, for certain pictures, given a mark using below equation：
Wherein, V represents the marking result of single picture（Between 0.0 to 1.0）.I=1,2,3 represent the browsing time respectively, drag number of times, scaling number of times. 
Represent the judge weights of other three browse operation factors. 
Represent the normalized data of corresponding factor.H is definite value 1：When picture is not downloaded out-of-date, d=0 is made；When picture is downloaded out-of-date, d=1 is made, so ensures that the picture being downloaded scoring is 1 i.e. full marks, and the browsed picture do not downloaded can be according to
Standards of grading provide scoring：
For(i=1,2,3) we provide weights as follows：Collect browsing time, download time and scaling per pictures, drag number of times, form data matrix：
Wherein
Data after the use minimax value method normalization of the corresponding index of expression m pictures.Then, we seek data matrix F its covariance matrix C：
Then, we are by following equation, to calculate C characteristic value：
Wherein P is C characteristic vector,It is C characteristic value.This equation is solved, we can obtain 3 solutions, i.e.,：
We are right
It is ranked up, obtains a new disaggregation：
According to
In sortord, the order of our three column vectors to F and
Carry out corresponding
Sequence.According to many experiments data, we are corresponding to maximum
's
It is assigned to 0.5, second largest for 0.3, minimum is assigned to 0.2.Then just can basis
Computing formula, come calculate its scoring.
Marking result V has 2 points of effects.（1）All images in database are ranked up from high to low using V values, a general ranking results are obtained.（2）For a certain position specific user, he（She）Obviously it is only possible to browse to a part for all images.It is ranked up using the V values image browsed to it, filters out higher part of wherein giving a mark as the picture set for meeting its people's preference, relatively low part of giving a mark is used as the pictures for not meeting personal preference.In the same or similar image content of user's future searches, recorded according to existing personal preference image, to it, not browsed similar image is trained, and is obtained thinking the pictures for meeting its preference and is returned to user.
2. during personal like's picture is met for user's matching, we are extracted border, texture and the color characteristic of picture.Feature Extraction Technology scheme is as follows.For the ease of description, we take an image as explanation object, are designated as I.
（1）Boundary characteristic extraction
We are based on Sobel operators and carry out Boundary characteristic extraction.Because Sobel operators to noise have smoothing effect, so as to provide more accurate edge directional information.
First, I is switched into gray-scale map.I is read in a matrix type, is designated as i.The homography r that extracts I boundary image is handled the progress of i matrixes using Sobel operators.In r, limitrophe pixel represents that non-border pixel is black, and r is 0-1 matrixes, wherein 1 represents white, 0 represents black with white.
According to image size, r is divided into 16 rows × 16 and arranges totally 256 submatrixs by us, and each submatrix size is identical and not overlapping.Because possibly i matrixes can not be completely covered in 256 submatrixs, we are in partition process, and the size to submatrix is rounded downwards, finally gives up to fall matrix bottom and the redundance on right side.Give up the matrix fallen for this part, it is believed that because its respective pixel is located at image border, the boundary characteristic contribution for image can be ignored.
In each submatrix, our numbers of statistical elements 1 and the ratio of the total first prime number of submatrix.So, for picture I, we have obtained the characteristic vector of one 256 dimension, and every dimension of the vector represents the ratio that boundary pixel in a sub-pictures accounts for total pixel number.、
（2）Color feature extracted
For other models, color characteristic of the hsv models more suitable for extraction picture.We are using hsv models, it is necessary to extract the H of picture（hueForm and aspect）、S（saturationSaturation degree）、V（valueTone）Three features.Because the picture I tested is rgb models, so we are extracted I three primary colors feature and are respectively present in three matrixes first.Then hsv model conversions are carried out to I each pixel.
We obtain the hsv characteristic values in I on each pixel after converting, to h, s, the original matrix of tri- features of v is arranged, and it is merged into one 64（Wherein, h：32, v：16, s:16）The vectorial characteristic vector as I.
（3）Texture feature extraction
We employ gabor wave filters and carry out texture feature extraction.
First, it would be desirable to which i is divided into some square submatrixs, so that use gabor wave filters carry out feature extraction.In view of the image size and the accuracy of feature extraction tested, it is 32 × 32 that we, which take submatrix size, and each submatrix has the area of half submatrix adjacent thereto to overlap.For on the downside of i matrixes and right side can not quilt Matrix cover redundance, we using and Boundary characteristic extraction identical processing method, give up to fall this part matrix.Assuming that we have finally marked off N number of submatrix.
Gabor wave filters, wherein stage=4, orientation=6 are applied in each submatrix.Final each submatrix obtains the characteristic vector of one 48 dimension, and I pictures have obtained the characteristic vector of N number of 48 dimension.But different size picture N values are different, it is difficult to carry out ensuing image training.In order to solve the problem, while in view of the size of sample image, we extract 32 characteristic vectors to every pictures to be trained.In order that this 32 vectors are capable of the textural characteristics of representative image to greatest extent, we used following screening technique：First, 8 classes are divided into using kmeans algorithms to I N number of characteristic vector, and obtain the center vector C of each classi.The vectorial quantity that we take out in being calculated according to the ratio of the vectorial number of every class per class afterwards, and take out in every class respective numbers, apart from its center vector CiNearest vector.So, each pictures can obtain the eigenmatrix of one 32 dimension.
It should be noted that the center vector of each class is random acquirement during due to standard kmeans algorithm initializations, and our unpredictable every group of center vectors, this can cause the otherness of each classification results, so as to produce different training results.Therefore we have chosen the average effect trained several times as experimental result.
3. the technology for obtaining color of image, texture and boundary characteristic in accordance with the above, we extract meet user preference, do not meet user preference and the feature of test pictures collection (i.e. the not browsed image to be classified collection of user) after it is trained.The pictures that record meets user preference are G, do not meet user preference pictures for B, test pictures integrate as S.In order to carry out SVM training, G is identical with set B size.Concrete technical scheme is as follows：So that boundary characteristic is trained as an example, utilize above-mentioned Boundary characteristic extraction technology, picture in gathering G, B, S carries out Boundary characteristic extraction, obtain the respective eigenmatrix of G, B, S and carry out SVM training, obtain one and judge that image to be screened belongs to the grader for meeting preference or not meeting sets of preferences, filter out the picture for meeting user preference in S pictures using the grader.Color, textural characteristics training method are ibid.
Every pictures in test pictures collection are all manually demarcated, determine whether every image actually meets user preference.The test pictures collection calibration result trained by three kinds of modes above is carried out into computer with the actual calibration result of test pictures collection to compare, hit rate is obtained, gives rate performance indications for change.For in the training result of the different characteristic set of same group of test pictures collection, performance most preferably be to best suit user's requirement result, it is the combinations of features for being best suitable for such image set to select its corresponding combinations of features.
4. set up an account for belonging to personal for each user for having record, the searching preferences of relative users are recorded under the account, when user searches again for same keyword or similar keywords, according to existing user preference information, image searching result is classified and reorders that there is provided the picture close with its personal like to user.
5. when user searches again for same retrieval input picture or similar image, according to existing user preference information, image searching result is trained using the optimal characteristics set and new information of step 4, realize classification and reorder, the picture for being supplied to user close with its personal like, precision is improved constantly by such iterative process.Our method can rapidly restrain in fewer iterations number of times.
In order to examine the model that we are set up for the differentiation degree of accuracy of user preferences, the user that we let on is scored all pictures manually, is divided into satisfied or unsatisfied picture, and then the differentiation result again with our system is contrasted.Cross as mentioned before, group that is that picture is divided into user preferences according to 0.4 scoring threshold value by our system and not liking.Correct definition of accuracy of classifying is the ratio of the judicious picture number of system and picture sum.The accuracy data of six pictures is shown in Fig. 2.
In order to verify generally influence of the sample dominant characteristics to disaggregated model, we control other conditions constant and six pictures are tested in the presence of noise, and hit rate after being restrained by data processing, give rate for change.In order to examine user in our methods to search for the effect of the adjustment to result correctness repeatedly, we are used to verifying this validity with influence of the multiple search operation process of a pictures to hit rate.The concrete condition of experimental result is shown in Fig. 3.Experimental conditions show that our method can just reach stable precision after 4 iteration.For average hit rate and giving rate for change, experimental result is as shown in table 1：
Table 1
| Pictures | Sample size | Meet the picture number of user preferences | Meet the picture rate of user preferences | Hit rate | Give rate for change |
| Seabeach | 275 | 119 | 0.4327 | 0.7586 | 0.5546 |
| The earth | 199 | 95 | 0.4774 | 0.8550 | 0.6941 |
| Cloth | 200 | 100 | 0.5000 | 0.9901 | 1.0000 |
| Flower | 215 | 72 | 0.3348 | 0.7817 | 0.5278 |
| Sailing boat | 198 | 78 | 0.3939 | 0.8837 | 0.4871 |
| Ripple | 219 | 87 | 0.3973 | 0.8261 | 0.7191 |
As shown by data, our model is that have higher confidence level.
User browses quantity to the participation of picture every time, also influences whether the working effect of system.In experiment before, user browses 20% or so picture in each search procedure to same content.But this is not to be required for the fixed of system.In following experiment, we reduce user every time browse quantity.Fig. 4's test result indicate that, in the case of a small amount of picture is browsed even in user, system is by very little is influenceed, with higher robustness.
Claims (9)
1. the image collection sort method of the recessive operation behavior of a kind of digging user, it is characterised in that step is such as
Under：
Step 1：According to user when the content based on sample image carries out image retrieval the pictures that retrieve, with the picture browsing information of statistical tool counting user, including：Browsing time, download time and scaling, dragging number of times per pictures；
Step 2：After image statistics are obtained, for certain pictures, given a mark using below equation：
Wherein, V represents the marking result of single picture（Between 0.0 to 1.0）, i=1,2,3 represent the browsing time respectively, drag number of times, scaling number of times.
2.
Represent the judge weights of other three browse operation factors.
3.
Represent the normalized data of corresponding factor.
4.H is definite value 1：When picture is not downloaded out-of-date, d=0 is made；When picture is downloaded out-of-date, d=1 is made, so ensures that the picture being downloaded scoring is 1 i.e. full marks, and the browsed picture do not downloaded can be according to
Standards of grading provide scoring；（Formula is not complete, it is impossible to show）
For
(i=1,2,3) we provide weights as follows：Collect browsing time, download time and scaling per pictures, drag number of times, form data matrix：
Wherein
Data after the use minimax value method normalization of the corresponding index of expression m pictures.
5. then, we seek data matrix F its covariance matrix C：
Then, we are by following equation, to calculate C characteristic value：
Wherein P is C characteristic vector,
It is C characteristic value.
6. solving this equation, we can obtain 3 solutions, i.e.,：
We are right
It is ranked up, obtains a new disaggregation：
According to
In sortord, the order of our three column vectors to F andCarry out correspondingSequence.
7. according to many experiments data, we are corresponding to maximum
's
It is assigned to 0.5, second largest for 0.3, minimum is assigned to 0.2.
8. and then just can basis
Computing formula, come calculate its scoring；
According to appraisal result from high to low, the picture in pictures is divided into user preference picture subset and does not meet two subsets of user preference pictures（For a pictures, it is divided into according to its scoring V (I) using 0.4 for threshold value and meets preference and do not meet preference as standard）；
Step 3：It regard the subset of the not browsed similar image construction of user as test pictures collection in addition；
Step 4：Respectively to meeting user preference pictures, not meeting user preference pictures and three principal characters of test pictures collection extraction image：Boundary characteristic, textural characteristics and color characteristic, set up the eigenmatrix of each picture subset；
Step 5：After the eigenmatrix of picture subset is obtained, SVM training is carried out to the picture in test pictures collection first against three single features, obtain judging that the image in test pictures collection belongs to the grader for meeting user preference or not meeting user preference subset, and then obtain thinking the picture for meeting user preference；Three single features eigenmatrixes are merged into a total eigenmatrix again, SVM training is carried out to the picture in test pictures collection, obtains thinking the picture for meeting user preference；
Step 6：Every pictures in test pictures collection are all manually demarcated, determine whether every image actually meets user preference；
Step 7：Step will be passed through（5）Different characteristic combination the test pictures collection calibration result trained and step（6）The result manually demarcated is compared, and selectes the characteristic set for being best suitable for such pictures；
Step 8:Opened an account for each user for having record, image category and corresponding image feature information that its searching preferences, i.e. user searched for are recorded under the account.
9. when user searches again for same retrieval input picture or similar image, according to existing user preference information, utilize step（7）Optimal characteristics set and new information image searching result is trained, realize classification and reorder there is provided the picture close with its personal like to user, improve constantly precision by such iterative process.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310279399.7A CN103514254A (en) | 2013-07-04 | 2013-07-04 | Image set ordering method for mining hidden operation behavior |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310279399.7A CN103514254A (en) | 2013-07-04 | 2013-07-04 | Image set ordering method for mining hidden operation behavior |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103514254A true CN103514254A (en) | 2014-01-15 |
Family
ID=49896978
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310279399.7A Pending CN103514254A (en) | 2013-07-04 | 2013-07-04 | Image set ordering method for mining hidden operation behavior |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103514254A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103927348A (en) * | 2014-04-02 | 2014-07-16 | 华为技术有限公司 | Picture processing method, information acquiring method and device |
| WO2016000251A1 (en) * | 2014-07-04 | 2016-01-07 | Microsoft Technology Licensing, Llc | Personalized trending image search suggestion |
| CN106156063A (en) * | 2015-03-30 | 2016-11-23 | 阿里巴巴集团控股有限公司 | Correlation technique and device for object picture search results ranking |
| CN106814946A (en) * | 2017-01-13 | 2017-06-09 | 滁州昭阳电信通讯设备科技有限公司 | The control method and mobile terminal of a kind of picture browsing |
| CN107111826A (en) * | 2014-11-12 | 2017-08-29 | 谷歌公司 | Automatic selection of images applied |
| CN107209767A (en) * | 2015-12-29 | 2017-09-26 | 华为技术有限公司 | A kind of management method of multimedia file, electronic equipment and graphic user interface |
| CN109508321A (en) * | 2018-09-30 | 2019-03-22 | Oppo广东移动通信有限公司 | image display method and related product |
| CN110795579A (en) * | 2019-10-29 | 2020-02-14 | Oppo广东移动通信有限公司 | Image cleaning method, device, terminal and storage medium |
| CN111666438A (en) * | 2020-05-22 | 2020-09-15 | 东华大学 | Cloud photo album text keyword fuzzy search system and use method |
| CN112069337A (en) * | 2020-09-03 | 2020-12-11 | Oppo广东移动通信有限公司 | Image processing method, device, electronic device and storage medium |
| CN112084394A (en) * | 2020-09-09 | 2020-12-15 | 重庆广播电视大学重庆工商职业学院 | Search result recommendation method and device based on image recognition |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090307219A1 (en) * | 2008-06-05 | 2009-12-10 | Bennett James D | Image search engine using image analysis and categorization |
| US20110307491A1 (en) * | 2009-02-04 | 2011-12-15 | Fisk Charles M | Digital photo organizing and tagging method |
| US20120290589A1 (en) * | 2011-05-13 | 2012-11-15 | Canon Kabushiki Kaisha | Information processing apparatus, information processing method, and non-transitory computer readable storage medium |
| CN102799635A (en) * | 2012-06-27 | 2012-11-28 | 天津大学 | Image set ordering method driven by user |
| CN102855268A (en) * | 2011-06-03 | 2013-01-02 | 国际商业机器公司 | Image ranking method and system based on attribute correlation |
| CN102982165A (en) * | 2012-12-10 | 2013-03-20 | 南京大学 | Large-scale human face image searching method |
-
2013
- 2013-07-04 CN CN201310279399.7A patent/CN103514254A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090307219A1 (en) * | 2008-06-05 | 2009-12-10 | Bennett James D | Image search engine using image analysis and categorization |
| US20110307491A1 (en) * | 2009-02-04 | 2011-12-15 | Fisk Charles M | Digital photo organizing and tagging method |
| US20120290589A1 (en) * | 2011-05-13 | 2012-11-15 | Canon Kabushiki Kaisha | Information processing apparatus, information processing method, and non-transitory computer readable storage medium |
| CN102855268A (en) * | 2011-06-03 | 2013-01-02 | 国际商业机器公司 | Image ranking method and system based on attribute correlation |
| CN102799635A (en) * | 2012-06-27 | 2012-11-28 | 天津大学 | Image set ordering method driven by user |
| CN102982165A (en) * | 2012-12-10 | 2013-03-20 | 南京大学 | Large-scale human face image searching method |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103927348A (en) * | 2014-04-02 | 2014-07-16 | 华为技术有限公司 | Picture processing method, information acquiring method and device |
| US10459964B2 (en) | 2014-07-04 | 2019-10-29 | Microsoft Technology Licensing, Llc | Personalized trending image search suggestion |
| WO2016000251A1 (en) * | 2014-07-04 | 2016-01-07 | Microsoft Technology Licensing, Llc | Personalized trending image search suggestion |
| CN107111826A (en) * | 2014-11-12 | 2017-08-29 | 谷歌公司 | Automatic selection of images applied |
| CN106156063A (en) * | 2015-03-30 | 2016-11-23 | 阿里巴巴集团控股有限公司 | Correlation technique and device for object picture search results ranking |
| CN107209767A (en) * | 2015-12-29 | 2017-09-26 | 华为技术有限公司 | A kind of management method of multimedia file, electronic equipment and graphic user interface |
| CN106814946A (en) * | 2017-01-13 | 2017-06-09 | 滁州昭阳电信通讯设备科技有限公司 | The control method and mobile terminal of a kind of picture browsing |
| CN109508321A (en) * | 2018-09-30 | 2019-03-22 | Oppo广东移动通信有限公司 | image display method and related product |
| CN109508321B (en) * | 2018-09-30 | 2022-01-28 | Oppo广东移动通信有限公司 | Image display method and related product |
| CN110795579A (en) * | 2019-10-29 | 2020-02-14 | Oppo广东移动通信有限公司 | Image cleaning method, device, terminal and storage medium |
| CN110795579B (en) * | 2019-10-29 | 2022-11-18 | Oppo广东移动通信有限公司 | Picture cleaning method and device, terminal and storage medium |
| CN111666438A (en) * | 2020-05-22 | 2020-09-15 | 东华大学 | Cloud photo album text keyword fuzzy search system and use method |
| CN112069337A (en) * | 2020-09-03 | 2020-12-11 | Oppo广东移动通信有限公司 | Image processing method, device, electronic device and storage medium |
| CN112084394A (en) * | 2020-09-09 | 2020-12-15 | 重庆广播电视大学重庆工商职业学院 | Search result recommendation method and device based on image recognition |
| CN112084394B (en) * | 2020-09-09 | 2024-04-23 | 重庆广播电视大学重庆工商职业学院 | Search result recommending method and device based on image recognition |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103514254A (en) | Image set ordering method for mining hidden operation behavior | |
| CN102799635B (en) | The image collection sort method that a kind of user drives | |
| CN105512684B (en) | Logo automatic identifying method based on principal component analysis convolutional neural networks | |
| US6556710B2 (en) | Image searching techniques | |
| KR100997541B1 (en) | Method and device for user profile based image recommendation combining content based filtering and collaborative filtering for solving new product recommendation problem | |
| CN100357944C (en) | Image retrieving system, image classifying system, image retrieving program, image classifying program, image retrieving method and image classifying method | |
| US6522782B2 (en) | Image and text searching techniques | |
| US7545980B2 (en) | Method of and apparatus for classifying an image | |
| EP2615572A1 (en) | Image segmentation based on approximation of segmentation similarity | |
| JP2009251850A (en) | Commodity recommendation system using similar image search | |
| CN102779157B (en) | Method and device for searching images | |
| CN108305240A (en) | Picture quality detection method and device | |
| CN103164687B (en) | A kind of method and system of pornographic image detecting | |
| CN109471982B (en) | Web service recommendation method based on QoS (quality of service) perception of user and service clustering | |
| Romero et al. | Using complexity estimates in aesthetic image classification | |
| US6522780B1 (en) | Indexing of images and/or text | |
| CN115713755A (en) | Efficient and accurate image identification method for Spodoptera frugiperda | |
| CN108256473A (en) | Methods, devices and systems, computer readable storage medium for data amplification | |
| CN109460767A (en) | Rule-based convex print bank card number segmentation and recognition methods | |
| CN103324750A (en) | Method for personal screening of photo galleries on the basis of Bayesian network | |
| US6522779B2 (en) | Representing an image with a posterized joint histogram | |
| CN106204538B (en) | A kind of image partition method and system | |
| CN118277674A (en) | Personalized image content recommendation method based on big data analysis | |
| KR20110019117A (en) | Semantic based image search method | |
| CN104021220B (en) | Image Retrieval Method Based on Saliency Model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20140115 |