CN103064985A - Priori knowledge based image retrieval method - Google Patents
Priori knowledge based image retrieval method Download PDFInfo
- Publication number
- CN103064985A CN103064985A CN2013100330581A CN201310033058A CN103064985A CN 103064985 A CN103064985 A CN 103064985A CN 2013100330581 A CN2013100330581 A CN 2013100330581A CN 201310033058 A CN201310033058 A CN 201310033058A CN 103064985 A CN103064985 A CN 103064985A
- Authority
- CN
- China
- Prior art keywords
- image
- classification
- feature
- word
- characteristic
- 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.)
- Granted
Links
Images
Abstract
The invention relates to a priori knowledge based image retrieval method and belongs to the field of content based image retrieval. The method includes: using a measuring method for Euclidean distance added with dimensionality weight to eliminate a background portion in an image and only retain foreground content that a user concerns; and lowering the dimensionality weight of the background portion, and improving foreground dimensionality weight so as to reduce influences of background and increase influences, of foreground in the image, on retrieval results. The retrieval process includes: extracting local features of a target image, selecting corresponding dimensionality weight under a currently-comparing image type, utilizing the weighted Euclidean distance function for similarity measurement, and returning corresponding images according to the size of similarity. Compared with existing content based image retrieval methods, the priori knowledge based image retrieval method is more accurate in retrieval and higher in efficiency.
Description
Technical field
The present invention relates to a kind of image search method, particularly a kind of image search method of knowledge-based belongs to the CBIR field.
Background technology
Along with popularizing of computing machine and internet, developing rapidly of multimedia technology emerges a large amount of images, Voice ﹠ Video resource on the network.These have also proposed new research topic to people when facilitating to people and enriching people's cultural life: the resource that how to find out rapidly and accurately user's request from vast and numerous multimedia resource.
At present, image retrieval technologies commonly used is based on the image retrieval of content.In the CBIR field, characteristics of image tolerance is as the committed step of weighing image similarity, and the quality of final image result for retrieval is played vital effect.Therefore, good distance metric method of searching is significant for the research CBIR.Because image background varies, particularly deformation easily occurs in the structure of flexible article, and these all can have influence on the proper vector that proposes from image, carry out on this basis the retrieval precision that distance metric will reduce image.
The CBIR method at first is Characteristic of Image in the analysis image storehouse, and extraction can reflect the proper vector of this picture material, deposits in the corresponding feature database.When carrying out image retrieval, to each given test pattern, the analysis image feature is also extracted the proper vector that can reflect this picture material; Then the proper vector extracted and the proper vector in the feature database are mated, search in image library according to matching result, just can obtain the most similar with it image.Wherein the process of proper vector coupling often utilizes the distance between the proper vector to carry out similarity measurement, and distance commonly used has Euclidean distance, weighted euclidean distance, Ming Shi distance etc.
The advantage of these distance functions is that all parameters and data all are to set in advance or directly extract to obtain from image, therefore easy to use, calculate simple.But their shortcoming is obvious too: owing to not considering priori, result for retrieval tends to not fully up to expectations.
Present content-based characteristics of image retrieval all is difficult to obtain a balance in efficient and effect with Similarity Measuring Algorithm, and this problem is based on an important key issue in the field of image search of content.Because the existence of this problem has hindered the widespread use of CBIR.
Summary of the invention
The objective of the invention is to propose the image search method of knowledge-based in order to overcome the deficiency of existing characteristics of image measure existence.
The objective of the invention is to be achieved through the following technical solutions.
A kind of image search method of knowledge-based, its operation steps is as follows: comprise training process and test process.
Described training process comprises that step 1 to step 3, is specially:
Step 1, set up the visual vocabulary table.
Every class image of successively training data being concentrated carries out analyzing and processing, for every class image is set up a visual vocabulary table.The concrete steps that the image of a classification of use is set up the visual vocabulary table of this classification image are:
Step 1.1: obtain training data and concentrate each Characteristic of Image vector in such other image, with symbol (f
1, f
2..., f
n) expression, wherein, f
1, f
2..., f
nA feature of difference presentation video, the number of feature in the n representation feature vector; Described feature is multidimensional, and its dimension represents with N, N 〉=2.
The described method of obtaining every concentrated Characteristic of Image vector of training data comprises: surf(Speed-up Robust Features) algorithm and sift(Scale Invariant Feature Transform) algorithm.
Step 1.2: this classification Characteristic of Image vector (f that obtains for step 1.1
1, f
2..., f
n) in feature f
1, f
2..., f
nCarry out cluster, obtain m feature classification, m is positive integer;
Step 1.3: for m the feature classification that step 1.2 obtains, obtain other word of each feature class.Described other word of each feature class is the central point of whole proper vector clusters in this feature classification.
Step 1.4: the visual vocabulary table that obtains this classification image.The visual vocabulary table of described this classification image is comprised of other word of each feature class of this classification class image.
Step 2, on the basis of step 1 operation, add up the word distribution histogram of every class image.For a class image, the concrete operation step of adding up the word distribution histogram of this classification image is:
Step 2.1: set a statistic for each word in the visual vocabulary table of this classification image, use respectively symbol T
1, T
2..., T
mExpression, and T is set
1, T
2..., T
mInitial value be 0.
Step 2.2: use successively each the Characteristic of Image vector (f in this classification
1, f
2..., f
n) in each feature and the word in the visual vocabulary table of this classification image compare one by one, find out the shortest word of distance, and the statistic that this word is corresponding is from increasing 1.
Step 2.3: to statistic T
1, T
2..., T
mCarry out normalized.Be specially: use respectively statistic T
1, T
2..., T
mCount summation divided by this classification Characteristic of Image, the result who obtains uses respectively T '
1, T '
2..., T
m'.
Step 2.4: the word distribution histogram that obtains this classification image.The statistic T of described word distribution histogram after by normalization corresponding to each word in the visual vocabulary table of this classification image '
1, T '
2..., T
m' form.
Step 3, on the basis of step 2 operation, calculate each classification characteristics of image characteristic of correspondence weights.For the image of a classification, the computing method of these classification characteristics of image characteristic of correspondence weights are:
Step 3.1: find out T ' in the word distribution histogram of this classification image
1, T '
2..., T
m' in maximal value, use symbol T
Max' expression.
Step 3.2: at such Characteristic of Image vector (f
1, f
2..., f
n) in find M and step 3.1 to obtain T
MaxThe immediate feature of ' distance, the matrix of generation M * N, matrix symbol A
M * N, the M value is by artificial appointment, 10≤M≤200 and M≤n.
Step 3.3: compute matrix A
M * NIn the inverse of variance of every column data, use respectively symbol σ
1, σ
2..., σ
N
Step 3.4: to matrix A described in the step 3.3
M * NIn the σ reciprocal of variance of every column data
1, σ
2..., σ
NCarry out normalized, obtain this classification characteristics of image characteristic of correspondence weights, use respectively ω
iExpression, 1≤i≤N.Wherein,
Operation through above-mentioned steps obtains these classification characteristics of image characteristic of correspondence weights ω for each class image
i
Described test process comprises step 4, is specially:
Step 4, on the basis of step 3 operation, test pattern is retrieved.
Step 4.1: the proper vector of obtaining test pattern.
Step 4.2: each feature of every image that each feature of test pattern is concentrated with training data is successively carried out distance metric.Wherein, a certain feature (F represents with symbol) of test pattern concentrates the distance (d represents with symbol) of certain feature (G represents with symbol) of a certain image (I represents with symbol) can pass through formula (1) calculating with training data.
Wherein, f
iI dimension data for the feature F of test pattern; g
iConcentrate the i dimension data of the feature G of image I for training data; ω
iFeature G characteristic of correspondence weights for the concentrated image I of training data.
Step 4.3: use symbol P
kMatch point logarithm corresponding to every image that the expression training data is concentrated, its initial value is set as 0; Wherein, 1≤k≤K, K represent the quantity of the concentrated image of training data.Every image concentrating for training data, successively with a certain feature of test pattern respectively and the distance between each feature of this training image as one group of data, and from these group data, pick out minimum value and sub-minimum, use respectively symbol d
MinAnd d '
MinExpression.If d
Min/ d '
Min＜θ, 0.5＜θ＜0.9 is then with match point logarithm P
kFrom increasing 1.
Step 4.3: for every concentrated image of training data, according to its corresponding match point logarithm P
kOrder from big to small is to training image ordering and output.Match point logarithm P
kLarger, this match point logarithm P
kCorresponding training image is more similar to test pattern.
Operation through above-mentioned steps can realize the retrieval to test pattern.
Beneficial effect
The image search method of the knowledge-based that the present invention proposes is compared with existing CBIR method, the retrieving of the inventive method uses the measure of the Euclidean distance that adds the dimension weights, its background that derives from the field of video processing is rejected algorithm: namely the background parts in the image is rejected, only kept the foreground content that the user pays close attention to.In the algorithm dimension weights of background parts are reduced, and the dimension weights of prospect dimension are improved, reduced the impact of image background, the prospect in the increase image is on the impact of last result for retrieval, so that retrieval is more accurate, efficient is higher.
Description of drawings
Fig. 1 is the structural framing schematic diagram of the image search method of knowledge-based in the specific embodiment of the invention.
Embodiment
For technical scheme of the present invention better is described, below in conjunction with accompanying drawing, by 1 embodiment, the present invention will be further described.
Picture in the present embodiment is divided into 4 classes: camera, refrigerator, notebook computer, mobile phone.Every class training image has 400.Use the image search method of knowledge-based training image to be trained and test pattern is retrieved, its structural framing as shown in Figure 1, concrete operation step comprises training process and test process.
Described training process comprises that step 1 to step 3, is specially:
Step 1, set up the visual vocabulary table.
Every class image of successively training data being concentrated carries out analyzing and processing, for every class image is set up a visual vocabulary table.The concrete steps that the image of a classification of use is set up the visual vocabulary table of this classification image are:
Step 1.1: use the surf algorithm to obtain training data and concentrate each Characteristic of Image vector (f in such other image
1, f
2..., f
n).Proper vector (f
1, f
2..., f
n) in be characterized as 64 the dimension.
Step 1.2: this classification Characteristic of Image vector (f that obtains for step 1.1
1, f
2..., f
n) in feature f
1, f
2..., f
nCarry out cluster, obtain 200 feature classifications;
Step 1.3: for 200 feature classifications that step 1.2 obtains, obtain other word of each feature class.Described other word of each feature class is the mean value of whole proper vectors in this feature classification.
Step 1.4: the visual vocabulary table that obtains this classification image.The visual vocabulary table of described this classification image is comprised of other word of each feature class of this classification class image.
Step 2, on the basis of step 1 operation, add up the word distribution histogram of every class image.For a class image, the concrete operation step of adding up the word distribution histogram of this classification image is:
Step 2.1: for a statistic T set in each word in the visual vocabulary table of this classification image
1, T
2..., T
m, and T is set
1, T
2..., T
mInitial value be 0.
Step 2.2: use successively each the Characteristic of Image vector (f in this classification
1, f
2..., f
n) in each feature and the word in the visual vocabulary table of this classification image compare one by one, find out the shortest word of distance, and the statistic that this word is corresponding is from increasing 1.
Step 2.3: to statistic T
1, T
2..., T
mCarry out normalized.Be specially: use respectively statistic T
1, T
2..., T
mCount summation divided by this classification Characteristic of Image, the result who obtains uses respectively T '
1, T '
2..., T
m'.
Step 2.4: the word distribution histogram that obtains this classification image.The statistic T of described word distribution histogram after by normalization corresponding to each word in the visual vocabulary table of this classification image '
1, T '
2..., T
m' form.
Step 3, on the basis of step 2 operation, calculate each classification characteristics of image characteristic of correspondence weights.For the image of a classification, the computing method of these classification characteristics of image characteristic of correspondence weights are:
Step 3.1: find out T ' in the word distribution histogram of this classification image
1, T '
2..., T
m' in maximum of T
Max'.
Step 3.2: at such Characteristic of Image vector (f
1, f
2..., f
n) in find 50 T that obtain with step 3.1
MaxThe immediate feature of ' distance generates 50 * 64 matrix, matrix symbol A
50 * 64
Step 3.3: compute matrix A
M * NIn the σ reciprocal of variance of every column data
1, σ
2..., σ
N
Step 3.4: to matrix A described in the step 3.3
50 * 64In the σ reciprocal of variance of every column data
1, σ
2..., σ
64Carry out normalized, obtain this classification characteristics of image characteristic of correspondence weights, use respectively ω
iExpression, 1≤i≤64.Wherein,
Operation through above-mentioned steps obtains these classification characteristics of image characteristic of correspondence weights ω for each class image
i
Described test process comprises step 4, is specially:
Step 4, on the basis of step 3 operation, test pattern is retrieved.
Step 4.1: use the surf algorithm to obtain the proper vector of test pattern, test pattern is a picture of mobile telephone.
Step 4.2: each feature of every image that each feature of test pattern is concentrated with training data is successively carried out distance metric.Wherein, certain feature G's of concentrated a certain the image I of a certain feature F of test pattern and training data can pass through formula (1) calculating apart from d.
Step 4.3: use symbol P
kMatch point logarithm corresponding to every image that the expression training data is concentrated, its initial value is set as 0.Every image concentrating for training data, successively with a certain feature of test pattern respectively and the distance between each feature of this training image as one group of data, and from these group data, pick out minimum value and sub-minimum, use respectively symbol d
MinAnd d '
MinExpression.If d
Min/ d '
Min＜0.7,0.5＜θ＜0.9 is then with match point logarithm P
kFrom increasing 1.
Step 4.3: for every concentrated image of training data, according to its corresponding match point logarithm P
kOrder from big to small is to training image ordering and output.Match point logarithm P
kLarger, this match point logarithm P
kCorresponding training image is more similar to test pattern.
Operation through above-mentioned steps can realize the retrieval to test pattern.
Above-described specific descriptions; purpose, technical scheme and beneficial effect to invention further describe; institute is understood that; the above only is specific embodiments of the invention; be used for explaining the present invention, the protection domain that is not intended to limit the present invention, within the spirit and principles in the present invention all; any modification of making, be equal to replacement, improvement etc., all should be included within protection scope of the present invention.
Claims (2)
1. the image search method of a knowledge-based, it is characterized in that: its operation steps is as follows: comprise training process and test process;
Described training process comprises that step 1 to step 3, is specially:
Step 1, set up the visual vocabulary table;
Every class image of successively training data being concentrated carries out analyzing and processing, for every class image is set up a visual vocabulary table; The concrete steps that the image of a classification of use is set up the visual vocabulary table of this classification image are:
Step 1.1: obtain training data and concentrate each Characteristic of Image vector in such other image, with symbol (f
1, f
2..., f
n) expression, wherein, f
1, f
2..., f
nA feature of difference presentation video, the number of feature in the n representation feature vector; Described feature is multidimensional, and its dimension represents with N, N 〉=2;
Step 1.2: this classification Characteristic of Image vector (f that obtains for step 1.1
1, f
2..., f
n) in feature f
1, f
2..., f
nCarry out cluster, obtain m feature classification, m is positive integer;
Step 1.3: for m the feature classification that step 1.2 obtains, obtain other word of each feature class; Described other word of each feature class is the central point of whole proper vector clusters in this feature classification;
Step 1.4: the visual vocabulary table that obtains this classification image; The visual vocabulary table of described this classification image is comprised of other word of each feature class of this classification class image;
Step 2, on the basis of step 1 operation, add up the word distribution histogram of every class image; For a class image, the concrete operation step of adding up the word distribution histogram of this classification image is:
Step 2.1: set a statistic for each word in the visual vocabulary table of this classification image, use respectively symbol T
1, T
2..., T
mExpression, and T is set
1, T
2..., T
mInitial value be 0;
Step 2.2: use successively each the Characteristic of Image vector (f in this classification
1, f
2..., f
n) in each feature and the word in the visual vocabulary table of this classification image compare one by one, find out the shortest word of distance, and the statistic that this word is corresponding is from increasing 1;
Step 2.3: to statistic T
1, T
2..., T
mCarry out normalized; Be specially: use respectively statistic T
1, T
2..., T
mCount summation divided by this classification Characteristic of Image, the result who obtains uses respectively T '
1, T '
2..., T
m';
Step 2.4: the word distribution histogram that obtains this classification image; The statistic T of described word distribution histogram after by normalization corresponding to each word in the visual vocabulary table of this classification image '
1, T '
2..., T
m' form;
Step 3, on the basis of step 2 operation, calculate each classification characteristics of image characteristic of correspondence weights; For the image of a classification, the computing method of these classification characteristics of image characteristic of correspondence weights are:
Step 3.1: find out T ' in the word distribution histogram of this classification image
1, T '
2..., T
m' in maximal value, use symbol T
Max' expression;
Step 3.2: at such Characteristic of Image vector (f
1, f
2..., f
n) in find M and step 3.1 to obtain T
MaxThe immediate feature of ' distance, the matrix of generation M * N, matrix symbol A
M * N, the M value is by artificial appointment, 10≤M≤200 and M≤n;
Step 3.3: compute matrix A
M * NIn the inverse of variance of every column data, use respectively symbol σ
1, σ
2..., σ
N
Step 3.4: to matrix A described in the step 3.3
M * NIn the σ reciprocal of variance of every column data
1, σ
2..., σ
NCarry out normalized, obtain this classification characteristics of image characteristic of correspondence weights, use respectively ω
iExpression, 1≤i≤N; Wherein,
Operation through above-mentioned steps obtains these classification characteristics of image characteristic of correspondence weights ω for each class image
i
Described test process comprises step 4, is specially:
Step 4, on the basis of step 3 operation, test pattern is retrieved;
Step 4.1: the proper vector of obtaining test pattern;
Step 4.2: each feature of every image that each feature of test pattern is concentrated with training data is successively carried out distance metric; Wherein, a certain feature F of test pattern represent with training data concentrate a certain image I certain feature G can pass through formula (1) calculating apart from d;
Wherein, f
iI dimension data for the feature F of test pattern; g
iConcentrate the i dimension data of the feature G of image I for training data; ω
iFeature G characteristic of correspondence weights for the concentrated image I of training data;
Step 4.3: use symbol P
kMatch point logarithm corresponding to every image that the expression training data is concentrated, its initial value is set as 0; Wherein, 1≤k≤K, K represent the quantity of the concentrated image of training data; Every image concentrating for training data, successively with a certain feature of test pattern respectively and the distance between each feature of this training image as one group of data, and from these group data, pick out minimum value and sub-minimum, use respectively symbol d
MinAnd d '
MinExpression; If d
Min/ d '
Min＜θ, 0.5＜θ＜0.9 is then with match point logarithm P
kFrom increasing 1;
Step 4.3: for every concentrated image of training data, according to its corresponding match point logarithm P
kOrder from big to small is to training image ordering and output; Match point logarithm P
kLarger, this match point logarithm P
kCorresponding training image is more similar to test pattern;
Operation through above-mentioned steps can realize the retrieval to test pattern.
2. the image search method of a kind of knowledge-based as claimed in claim 1 is characterized in that: the method for obtaining every Characteristic of Image vector that training data concentrates described in its step 1 step 1.1 comprises: surf algorithm and sift algorithm.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310033058.1A CN103064985B (en) | 2013-01-28 | 2013-01-28 | Priori knowledge based image retrieval method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310033058.1A CN103064985B (en) | 2013-01-28 | 2013-01-28 | Priori knowledge based image retrieval method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103064985A true CN103064985A (en) | 2013-04-24 |
| CN103064985B CN103064985B (en) | 2015-07-22 |
Family
ID=48107615
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310033058.1A Expired - Fee Related CN103064985B (en) | 2013-01-28 | 2013-01-28 | Priori knowledge based image retrieval method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103064985B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104504049A (en) * | 2014-12-20 | 2015-04-08 | 辽宁师范大学 | Retrieval method of color images based on quaternion Harmonic-Fourier moments |
| CN104867026A (en) * | 2014-02-21 | 2015-08-26 | 北京京东尚科信息技术有限公司 | Method and system for providing goods image and terminal device for outputting goods image |
| CN105469096A (en) * | 2015-11-18 | 2016-04-06 | 南京大学 | Feature bag image retrieval method based on Hash binary code |
| CN106844785A (en) * | 2017-03-15 | 2017-06-13 | 浙江工业大学 | A kind of CBIR method based on conspicuousness segmentation |
| WO2017143979A1 (en) * | 2016-02-22 | 2017-08-31 | 中兴通讯股份有限公司 | Image search method and device |
| CN109101602A (en) * | 2018-08-01 | 2018-12-28 | 腾讯科技（深圳）有限公司 | Image encrypting algorithm training method, image search method, equipment and storage medium |
| CN111125398A (en) * | 2019-12-19 | 2020-05-08 | 云粒智慧科技有限公司 | Picture information retrieval method, device, equipment and medium |
| CN113392257A (en) * | 2021-06-23 | 2021-09-14 | 泰康保险集团股份有限公司 | Image retrieval method and device |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040091137A1 (en) * | 2002-11-04 | 2004-05-13 | Samsung Electronics Co., Ltd. | System and method for detecting face |
| CN102208038A (en) * | 2011-06-27 | 2011-10-05 | 清华大学 | Image classification method based on visual dictionary |
| CN102332092A (en) * | 2011-09-14 | 2012-01-25 | 广州灵视信息科技有限公司 | Flame detection method based on video analysis |
-
2013
- 2013-01-28 CN CN201310033058.1A patent/CN103064985B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040091137A1 (en) * | 2002-11-04 | 2004-05-13 | Samsung Electronics Co., Ltd. | System and method for detecting face |
| CN102208038A (en) * | 2011-06-27 | 2011-10-05 | 清华大学 | Image classification method based on visual dictionary |
| CN102332092A (en) * | 2011-09-14 | 2012-01-25 | 广州灵视信息科技有限公司 | Flame detection method based on video analysis |
Non-Patent Citations (1)
| Title |
|---|
| 周媛等: "基于内容的多特征图像检索中的关键技术", 《图书馆》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104867026A (en) * | 2014-02-21 | 2015-08-26 | 北京京东尚科信息技术有限公司 | Method and system for providing goods image and terminal device for outputting goods image |
| CN104867026B (en) * | 2014-02-21 | 2021-04-30 | 北京京东尚科信息技术有限公司 | Method and system for providing commodity image and terminal device for outputting commodity image |
| CN104504049A (en) * | 2014-12-20 | 2015-04-08 | 辽宁师范大学 | Retrieval method of color images based on quaternion Harmonic-Fourier moments |
| CN105469096A (en) * | 2015-11-18 | 2016-04-06 | 南京大学 | Feature bag image retrieval method based on Hash binary code |
| CN105469096B (en) * | 2015-11-18 | 2018-09-25 | 南京大学 | A kind of characteristic bag image search method based on Hash binary-coding |
| WO2017143979A1 (en) * | 2016-02-22 | 2017-08-31 | 中兴通讯股份有限公司 | Image search method and device |
| CN106844785A (en) * | 2017-03-15 | 2017-06-13 | 浙江工业大学 | A kind of CBIR method based on conspicuousness segmentation |
| CN109101602A (en) * | 2018-08-01 | 2018-12-28 | 腾讯科技（深圳）有限公司 | Image encrypting algorithm training method, image search method, equipment and storage medium |
| CN109101602B (en) * | 2018-08-01 | 2023-09-12 | 腾讯科技（深圳）有限公司 | Image retrieval model training method, image retrieval method, device and storage medium |
| CN111125398A (en) * | 2019-12-19 | 2020-05-08 | 云粒智慧科技有限公司 | Picture information retrieval method, device, equipment and medium |
| CN113392257A (en) * | 2021-06-23 | 2021-09-14 | 泰康保险集团股份有限公司 | Image retrieval method and device |
| CN113392257B (en) * | 2021-06-23 | 2023-06-16 | 泰康保险集团股份有限公司 | Image retrieval method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103064985B (en) | 2015-07-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103064985B (en) | Priori knowledge based image retrieval method | |
| WO2020108608A1 (en) | Search result processing method, device, terminal, electronic device, and storage medium | |
| CN101247470B (en) | Method realized by computer for detecting scene boundaries in videos | |
| US10664719B2 (en) | Accurate tag relevance prediction for image search | |
| Jiang et al. | Columbia-UCF TRECVID2010 Multimedia Event Detection: Combining Multiple Modalities, Contextual Concepts, and Temporal Matching. | |
| US9229956B2 (en) | Image retrieval using discriminative visual features | |
| CN102549603B (en) | Relevance-based image selection | |
| CN104881458B (en) | A kind of mask method and device of Web page subject | |
| Li et al. | Harvesting social images for bi-concept search | |
| CN110472652B (en) | Small sample classification method based on semantic guidance | |
| Yasmin et al. | Content based image retrieval by shape, color and relevance feedback | |
| CN108959305A (en) | A kind of event extraction method and system based on internet big data | |
| CN110362678A (en) | A kind of method and apparatus automatically extracting Chinese text keyword | |
| CN109086375A (en) | A kind of short text subject extraction method based on term vector enhancing | |
| CN103631932A (en) | Method for detecting repeated video | |
| CN103440262A (en) | Image searching system and image searching method basing on relevance feedback and Bag-of-Features | |
| CN111782833A (en) | Fine-grained cross-media retrieval method based on multi-model network | |
| CN104199838B (en) | A kind of user model constructing method based on label disambiguation | |
| Ma et al. | Spatial-content image search in complex scenes | |
| WO2020135054A1 (en) | Method, device and apparatus for video recommendation and storage medium | |
| CN107527058A (en) | A kind of image search method based on weighting local feature Aggregation Descriptor | |
| CN103279581A (en) | Method for performing video retrieval by compact video theme descriptors | |
| Mironica et al. | Fisher kernel based relevance feedback for multimodal video retrieval | |
| JP6017277B2 (en) | Program, apparatus and method for calculating similarity between contents represented by set of feature vectors | |
| CN104765764A (en) | Indexing method based on large-scale image |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20150722 Termination date: 20160128 |
|
| EXPY | Termination of patent right or utility model |