CN103246688A - Semantic hierarchy model image classification management method based on salient region sparse representation - Google Patents
Semantic hierarchy model image classification management method based on salient region sparse representation Download PDFInfo
- Publication number
- CN103246688A CN103246688A CN2012105048525A CN201210504852A CN103246688A CN 103246688 A CN103246688 A CN 103246688A CN 2012105048525 A CN2012105048525 A CN 2012105048525A CN 201210504852 A CN201210504852 A CN 201210504852A CN 103246688 A CN103246688 A CN 103246688A
- Authority
- CN
- China
- Prior art keywords
- image
- semantic
- label
- node
- level
- 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
- 238000007726 management method Methods 0.000 title claims abstract description 29
- 238000000034 method Methods 0.000 claims abstract description 44
- 230000008569 process Effects 0.000 claims abstract description 16
- 238000001514 detection method Methods 0.000 claims abstract description 7
- 238000012360 testing method Methods 0.000 claims description 27
- 238000012549 training Methods 0.000 claims description 24
- 230000008859 change Effects 0.000 claims description 12
- 239000004744 fabric Substances 0.000 claims description 12
- 239000000284 extract Substances 0.000 claims description 10
- 238000000605 extraction Methods 0.000 claims description 5
- 230000008520 organization Effects 0.000 claims description 5
- 230000005764 inhibitory process Effects 0.000 claims description 3
- 239000000203 mixture Substances 0.000 claims description 3
- 238000000638 solvent extraction Methods 0.000 claims description 3
- 230000019771 cognition Effects 0.000 abstract description 6
- 239000000463 material Substances 0.000 abstract description 5
- 238000010276 construction Methods 0.000 abstract description 2
- 230000000007 visual effect Effects 0.000 description 8
- 238000010801 machine learning Methods 0.000 description 3
- 230000008447 perception Effects 0.000 description 3
- 210000004556 brain Anatomy 0.000 description 2
- 238000003066 decision tree Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000003909 pattern recognition Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 241000282326 Felis catus Species 0.000 description 1
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 238000009412 basement excavation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000035479 physiological effects, processes and functions Effects 0.000 description 1
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a semantic hierarchy model image classification management method based on salient region sparse representation. The invention integrates the obvious detection model into the construction process of the image management model: firstly, constructing a hierarchical semantic annotation tree (HSLT) on the basis of an image salient region; then, the HSLT is used for carrying out hierarchical semantic annotation on the image, so that the cognition of the high-level semantic of the image is obtained; and finally, establishing an image level management model by taking the level semantic annotation information as a basis. The invention accords with the basic idea of managing images by people, effectively organizes and manages the images according to semantic information, vividly simulates the process of managing and storing the images by people, reduces the consumption of manpower, material resources, time and the like, and has important practical significance.
Description
Technical field
The present invention relates to the semantic hierarchies model image sort management method based on the marking area rarefaction representation that a kind of visually-perceptible combines with the concept perception, can be used for technical fields such as mass data classification, high-level semantic mark, image information retrieval.
Background technology
Along with the fast development of image digitization information acquisition apparatus and technology, how more a focus that becomes in image storage and the shared research is organized, browses and retrieved to science to these large nuber of images information more easily.Human visual system and brain cognitive process can extract the succinct useful information in the large nuber of images information fast, make up the semantic hierarchies model of image information, therefore imitate the mechanism of human vision information extraction and extraction of semantics, carry out one that the automatic Classification Management of image becomes in the computer vision research field and have challenging studying a question.Traditional image information storage Classification Management adopts manual sort's method to carry out generally according to filename and date mostly.And that manual sort's method is influenced by people's subjective factor is very big, and expends a large amount of time and efforts, inefficiency.Layer management is a natural thinking of large nuber of images information management, and existing image information layer management model roughly is divided three classes: 1) based on the hierarchical model of language text; 2) based on the hierarchical model of low layer visual properties; 3) based on the hierarchical model of semantic visual feature.
Hierarchical model based on language text is the hierarchical network structure that makes up on the basis of image information text label inner link, the text label here comprise artificial for the image definition and be image labeling automatically with the method for machine learning.Comparatively be typically word network (WordNet) model in this class model, from the inner link between the semantic angle excavation text vocabulary of word, thereby the structure hierarchical relationship can be used in field of image search, also can be used for auxiliary object identification.The hierarchical model of this class has great importance for the organization and administration image, but has but ignored the contact between the different images simultaneously.Hierarchical model based on the low layer visual signature is early stage model, is classification foundation with the low-level feature of image, and as color, shape, texture etc., the image with same low-layer visual signature is assigned in the same class.Each class node is chosen different low-level features as classification foundation in such hierarchical model, adopt the method for pattern-recognition or machine learning that characteristic node is organized into hierarchical structure for the feature of choosing, that adopt as Sivi is hLDA(hierarchy Latent Dirichlet Allocation) method make up hierarchical structure, Griffin adopts space pyramid coupling (Spatial Pyramid Matching) method study to be trained to required hierarchical model.
Although the two kinds of methods in front have obtained paying close attention to widely, but in a lot of actual application, receive very big restriction, be confined to be not enough to satisfy people's classificating requirement with the outward appearance visual properties of the image classification foundation as image based on the hierarchical model of low-level feature.Though the hierarchical model based on language text can carry out the level classification to image according to semantic information, yet also an evaluation criterion can not arranged in the process that makes up model, it is bigger influenced by artificial subjective factor.In addition, because concept is semantic and perception is semantic often also uncorrelated, for application such as image information management, mark, can not set up enough good hierarchical model.Low at front two kinds of image Classification Managements model efficiency, problems such as accuracy rate is low, adopt machine learning method to obtain the text label related with picture material automatically, and then be to use according to the method that makes up the semantic visual level to give birth to the text label, in computer vision in 2010 and pattern-recognition international conference (CVPR), people such as Li adopt improved LDA method to obtain the level semantic label of image, and then according to label with the image storage administration of classifying, obtained meeting human thinking's image management model.People such as Bannour proposed the semantic hierarchies construction method based on image, semantic in 2012.
Human brain is to set up on the basis of image cognition to image Classification Management model, and the human visual system provides the physiology foundation for people to the elementary cognition of image.Vision significance is a kind of method for expressing to the image perception, so marking area provided the expression of the elementary cognition of image, has reflected the concern of human vision to picture material, can influence people to a great extent to the cognition of image.
Summary of the invention
For addressing the above problem, the invention provides a kind of semantic hierarchies model image sort management method based on the marking area rarefaction representation, the process of image is deposited in the simulation people management of comparison image, has reduced consumption such as human and material resources and time, has important practical significance.
The apprizing system of sample treatment plant reaches above-mentioned technique effect for realizing above-mentioned technical purpose, and the present invention is achieved through the following technical solutions:
A kind of semantic hierarchies model image sort management method based on the marking area rarefaction representation may further comprise the steps:
The foundation of step 1) marking area
Source images is carried out definition judgment with remarkable object detection method to image, when the poor definition different time that exists in the image between the subregion, then extract the marking area of image with non-clear area inhibition method; Do not have the poor definition different time in image, the remarkable figure of the remarkable detection model computed image that merges with many features then extracts the marking area of image then according to remarkable figure;
Step 2) the image, semantic label obtains
Content to marking area is carried out automatic semantic tagger with the STF method, obtains the most important semantic topic label of image and the semantic information collection of image marking area automatically;
Step 3) is set up the image level semanteme
In the training stage, suppose that the training set image collection is
, at first need the every width of cloth image in the training set
, being retrieved as its of location by the semantic topic label can represent
The semantic topic label of main semantic information
, obtain image-tag set
Need manually for image in the training set then
Define under it
The level label
, so just can obtain the set of image-level semantic label
,
...,
; At last, the thought training of image-layer semantic label being gathered with nCRP is configured to level semantic tagger tree HSLT;
At test phase, at first with test set
In every width of cloth image
Utilize the semantic label obtaining step extract with
Corresponding remarkable semantic information collection
, obtain the image corresponding with test set-significantly semantic information collection set
The HSLT that generates with the training stage is to every width of cloth test pattern then
Carry out the level mark, current test pattern
Subject categories be
, significantly the semantic information collection is
, right
It is as follows to carry out level mark concrete steps:
Step1: be present image
Create semantic path collection
And put sky, get set
In the label classification
Step2: scan each leaf node of HSLT successively, if there is leaf node
Subject categories and current taking-up
Corresponding, then stop scanning, change Step3; If do not exist, illustrate that then the test set image coverage of HSLT is not enough, will
The level label be set to
, change Step5;
Step3: from what navigate to
Calculating its contrary semantic path is
Step4: with the contrary semantic path counter-rotating that obtains among the Step3, thereby obtain
The semantic path of level
=
Step5: with current acquisition
Put into set
, if current
Set is not empty, continues to get wherein next label
, change Step2; If it is current
Be sky, then change Step6;
Step6: ask for the gained set of paths
The longest public semantic path
, be present image
Semantic path, also be
Level semantic tagger information;
Step7: finish;
Step 4) image, semantic layer management model
All set up at each node layer except leaf node " ... mix " file, and under root directory, set up " other " file simultaneously;
For setting up good administrative model, need fill to realize that to it it is to the function of image library organization and administration with all training sets and test set picture; In filling the model process, level semantic tagger information according to every width of cloth image correspondence, it is nearest public semantic path, root node from administrative model, select corresponding branch child node according to the semantic information of next level, up to arriving the semantics folder that the longest semantic path most end node is represented, if the semantics folder of least significant end is the non-leaf node of HSLT tree, then image is deposited in the mixed file folder under this semantics folder, remaining deposits image in this document folder down.
Further, when image comprises multiple semantic information, choose a kind of therein or a few effectively the method for the theme of identification image classification is as follows:
Because the significantly influence of figure partitioning algorithm, the marking area that is partitioned into might comprise the background area of part, when perhaps comprising multiple well-marked target and cause the variation of remarkable semantic information owing to image itself, a plurality of candidates' subject categories may appear after to marking area piece mark through the STF algorithm; Suppose source images
, through handling its marking area of extraction be
, right then
Carry out semantic tagger with STF, after supposing to mark
In comprise
Individual candidate's semantic label, the semantic label set is designated as
, calculate by (1)
The probability that belongs to each theme label
Wherein,
Expression be pixel in the image,
The pixel set of corresponding conditions, then image are satisfied in representative
Topmost semantic topic label
Can determine by (2);
Often have a plurality of targets in the image in the marking area of attractive attention, the corresponding a plurality of semantic informations of a plurality of targets form remarkable semantic information set, thereby influence final classification results, therefore need be image
Defining one comprises
The tag set of individual remarkable semantic information
,
Be
Subclass, be defined as (3);
Wherein,
For the control parameter, when image belongs to classification
Probability greater than belonging to the semantic topic classification
Probability
The time, think
Be significant semantic information, there is the final classification that can influence image in it, will
Put into image
The set of remarkable semantic information
In.
Further, described semantic path refers to that the root node from HSLT begins, and walks toward next node layer of tree successively according to the sensing of certain child nodes of root node, till the leaf node of HSLT, records the node semantic information of its process.
Further, described contrary semantic path is the leaf node from HSLT, passes through the father node of node successively, till root node, the semantic information of the node that records on the path of process.
Further, described nearest public parent refers to common parent nearest on the contrary path of two or more a plurality of classes.
The invention has the beneficial effects as follows:
1, the present invention meets the basic concept of people's managing image, and according to the effective organization and administration image of semantic information, the process of image is deposited in the simulation people of comparison image management, has reduced consumption such as human and material resources and time, has important practical significance;
2, after hierarchical model is populated by image, namely when image storing classifiedly under the current level file finish after, this model also has great importance for follow-up operations such as retrieval, because this model is based on semantic information, when needs are searched the image of a certain classification, just can retrieve by level according to the classification under this classification is on certain level, suppose that the image library size is
, when not according to hierarchical model, the time complexity of retrieving certain image is
, the complexity in the layer management model then is
, can shorten retrieval time greatly, improved effectiveness of retrieval, have important and practical meanings.
Embodiment
A kind of semantic hierarchies model image sort management method based on the marking area rarefaction representation may further comprise the steps:
The foundation of step 1) marking area
Source images is carried out definition judgment with remarkable object detection method to image, when the poor definition different time that exists in the image between the subregion, then extract the marking area of image with non-clear area inhibition method; Do not have the poor definition different time in image, the remarkable figure of the remarkable detection model computed image that merges with many features then extracts the marking area of image then according to remarkable figure;
Step 2) the image, semantic label obtains
Content to marking area is carried out automatic semantic tagger with the STF method, obtains the most important semantic topic label of image and the semantic information collection of image marking area automatically;
Semantic Texton Forests (STF) is a kind of new efficient semanteme marking method based on the image low-level feature.This method is based upon on the decision-tree model, directly the pixel of image is carried out semantic tagger as object of classification, compares with additive method, and this method can effectively reduce calculated amount.Because adopt the method for decision tree, this method embodies good jump in training and testing, demonstrated fully the hierarchical clustering relation of node.What this method was chosen simultaneously is the feature of local rectangular area, has very big advantage aspect accuracy and the time performance than other method.
Step 3) is set up the image level semanteme
In the training stage, suppose that the training set image collection is
, at first need the every width of cloth image in the training set
, being retrieved as its of location by the semantic topic label can represent
The semantic topic label of main semantic information
, obtain image-tag set
Need manually for image in the training set then
Define under it
The level label
, so just can obtain the set of image-level semantic label
,
...,
; At last, the thought training of image-layer semantic label being gathered with nCRP is configured to level semantic tagger tree HSLT;
At test phase, at first with test set
In every width of cloth image
Utilize the semantic label obtaining step extract with
Corresponding remarkable semantic information collection
, obtain the image corresponding with test set-significantly semantic information collection set
The HSLT that generates with the training stage is to every width of cloth test pattern then
Carry out the level mark, current test pattern
Subject categories be
, significantly the semantic information collection is
, right
It is as follows to carry out level mark concrete steps:
Step1: be present image
Create semantic path collection
And put sky, get set
In the label classification
Step2: scan each leaf node of HSLT successively, if there is leaf node
Subject categories and current taking-up
Corresponding, then stop scanning, change Step3; If do not exist, illustrate that then the test set image coverage of HSLT is not enough, will
The level label be set to
, change Step5;
Step3: from what navigate to
Calculating its contrary semantic path is
Step4: with the contrary semantic path counter-rotating that obtains among the Step3, thereby obtain
The semantic path of level
=
Step5: with current acquisition
Put into set
, if current
Set is not empty, continues to get wherein next label
, change Step2; If it is current
Be sky, then change Step6;
Step6: ask for the gained set of paths
The longest public semantic path
, be present image
Semantic path, also be
Level semantic tagger information;
Step7: finish;
Step 4) image, semantic layer management model
All set up at each node layer except leaf node " ... mix " file, and under root directory, set up " other " file simultaneously;
For setting up good administrative model, need fill to realize that to it it is to the function of image library organization and administration with all training sets and test set picture; In filling the model process, level semantic tagger information according to every width of cloth image correspondence, it is nearest public semantic path, root node from administrative model, select corresponding branch child node according to the semantic information of next level, up to arriving the semantics folder that the longest semantic path most end node is represented, if the semantics folder of least significant end is the non-leaf node of HSLT tree, then image is deposited in the mixed file folder under this semantics folder, remaining deposits image in this document folder down.
Further, when image comprises multiple semantic information, choose a kind of therein or a few effectively the method for the theme of identification image classification is as follows:
Because the significantly influence of figure partitioning algorithm, the marking area that is partitioned into might comprise the background area of part, when perhaps comprising multiple well-marked target and cause the variation of remarkable semantic information owing to image itself, a plurality of candidates' subject categories may appear after to marking area piece mark through the STF algorithm; Suppose source images
, through handling its marking area of extraction be
, right then
Carry out semantic tagger with STF, after supposing to mark
In comprise
Individual candidate's semantic label, the semantic label set is designated as
, calculate by (1)
The probability that belongs to each theme label
Wherein,
Expression be pixel in the image,
The pixel set of corresponding conditions, then image are satisfied in representative
Topmost semantic topic label
Can determine by (2);
Often have a plurality of targets in the image in the marking area of attractive attention, the corresponding a plurality of semantic informations of a plurality of targets form remarkable semantic information set, thereby influence final classification results, therefore need be image
Defining one comprises
The tag set of individual remarkable semantic information
,
Be
Subclass, be defined as (3);
Wherein,
For the control parameter, when image belongs to classification
Probability greater than belonging to the semantic topic classification
Probability
The time, think
Be significant semantic information, there is the final classification that can influence image in it, will
Put into image
The set of remarkable semantic information
In.
Described semantic path refers to that the root node from HSLT begins, and walks toward next node layer of tree successively according to the sensing of certain child nodes of root node, till the leaf node of HSLT, records the node semantic information of its process.
Described contrary semantic path is the leaf node from HSLT, passes through the father node of node successively, till root node, the semantic information of the node that records on the path of process.
Described nearest public parent refers to common parent nearest on the contrary path of two or more a plurality of classes.
Two paths are supposed in the longest public semantic path
,
Leaf node be respectively
,
,
With
Nearest public parent be
, then
With
The longest public semantic path refer to from root node to
The path, in like manner be used for asking the longest public semantic path in a plurality of paths, single-pathway
The longest public semantic path be
Itself.For example the longest public semantic path of path " the biological animal cat of image library " and " the biological plant flowers of image library " is " image library biology ".
Because the STF during semantic label obtains is based on the supervised learning of test set, therefore, is image
The location
It is the subject categories that has defined.Equally, in the supervised learning process of HSLT, select the wider test set of coverage to construct and comprise the more comprehensive HSLT of classification, if selected test set can not be contained all semantic topic classifications, then may occur among the Step1 scanning less than with
The situation of corresponding leaf node, in this case, because the defective of training set, the HLST that current training obtains can not find
The path, caused the generation of error, the label that present image is set is
, the image of this class is divided into a class, remedy the indeterminable classification problem of HLST.In a word, by test process, can obtain test set level semantic information set
,
,
……,
}。
Claims (5)
1. based on the semantic hierarchies model image sort management method of marking area rarefaction representation, it is characterized in that, may further comprise the steps:
The foundation of step 1) marking area
Source images is carried out definition judgment with remarkable object detection method to image, when the poor definition different time that exists in the image between the subregion, then extract the marking area of image with non-clear area inhibition method; Do not have the poor definition different time in image, the remarkable figure of the remarkable detection model computed image that merges with many features then extracts the marking area of image then according to remarkable figure;
Step 2) the image, semantic label obtains
Content to marking area is carried out automatic semantic tagger with the STF method, obtains the most important semantic topic label of image and the semantic information collection of image marking area automatically;
Step 3) is set up the image level semanteme
In the training stage, suppose that the training set image collection is
, at first need the every width of cloth image in the training set
, being retrieved as its of location by the semantic topic label can represent
The semantic topic label of main semantic information
, obtain image-tag set
Need manually for image in the training set then
Define under it
The level label
, so just can obtain the set of image-level semantic label
,
...,
; At last, the thought training of image-layer semantic label being gathered with nCRP is configured to level semantic tagger tree HSLT;
At test phase, at first with test set
In every width of cloth image
Utilize the semantic label obtaining step extract with
Corresponding remarkable semantic information collection
, obtain the image corresponding with test set-significantly semantic information collection set
The HSLT that generates with the training stage is to every width of cloth test pattern then
Carry out the level mark, current test pattern
Subject categories be
, significantly the semantic information collection is
, right
It is as follows to carry out level mark concrete steps:
Step1: be present image
Create semantic path collection
And put sky, get set
In the label classification
Step2: scan each leaf node of HSLT successively, if there is leaf node
Subject categories and current taking-up
Corresponding, then stop scanning, change Step3; If do not exist, illustrate that then the test set image coverage of HSLT is not enough, will
The level label be set to
, change Step5;
Step3: from what navigate to
Calculating its contrary semantic path is
Step4: with the contrary semantic path counter-rotating that obtains among the Step3, thereby obtain
The semantic path of level
=
Step5: with current acquisition
Put into set
, if current
Set is not empty, continues to get wherein next label
, change Step2; If it is current
Be sky, then change Step6;
Step6: ask for the gained set of paths
The longest public semantic path
, be present image
Semantic path, also be
Level semantic tagger information;
Step7: finish;
Step 4) image, semantic layer management model
All set up at each node layer except leaf node " ... mix " file, and under root directory, set up " other " file simultaneously;
For setting up good administrative model, need fill to realize that to it it is to the function of image library organization and administration with all training sets and test set picture; In filling the model process, level semantic tagger information according to every width of cloth image correspondence, it is nearest public semantic path, root node from administrative model, select corresponding branch child node according to the semantic information of next level, up to arriving the semantics folder that the longest semantic path most end node is represented, if the semantics folder of least significant end is the non-leaf node of HSLT tree, then image is deposited in the mixed file folder under this semantics folder, remaining deposits image in this document folder down.
2. the semantic hierarchies model image sort management method based on the marking area rarefaction representation according to claim 1, it is characterized in that: when image comprises multiple semantic information, choose a kind of therein or a few effectively the method for the theme of identification image classification is as follows:
Because the significantly influence of figure partitioning algorithm, the marking area that is partitioned into might comprise the background area of part, when perhaps comprising multiple well-marked target and cause the variation of remarkable semantic information owing to image itself, a plurality of candidates' subject categories may appear after to marking area piece mark through the STF algorithm; Suppose source images
, through handling its marking area of extraction be
, right then
Carry out semantic tagger with STF, after supposing to mark
In comprise
Individual candidate's semantic label, the semantic label set is designated as
, calculate by (1)
The probability that belongs to each theme label
Wherein,
Expression be pixel in the image,
The pixel set of corresponding conditions, then image are satisfied in representative
Topmost semantic topic label
Can determine by (2);
Often have a plurality of targets in the image in the marking area of attractive attention, the corresponding a plurality of semantic informations of a plurality of targets form remarkable semantic information set, thereby influence final classification results, therefore need be image
Defining one comprises
The tag set of individual remarkable semantic information
,
Be
Subclass, be defined as (3);
(3)
Wherein,
For the control parameter, when image belongs to classification
Probability greater than belonging to the semantic topic classification
Probability
The time, think
Be significant semantic information, there is the final classification that can influence image in it, will
Put into image
The set of remarkable semantic information
In.
3. the semantic hierarchies model image sort management method based on the marking area rarefaction representation according to claim 1, it is characterized in that: described semantic path refers to that the root node from HSLT begins, sensing according to certain child nodes of root node is walked toward next node layer of tree successively, till the leaf node of HSLT, record the node semantic information of its process.
4. the semantic hierarchies model image sort management method based on the marking area rarefaction representation according to claim 1, it is characterized in that: described contrary semantic path is the leaf node from HSLT, pass through the father node of node successively, till root node, the semantic information of the node that records on the path of process.
5. the semantic hierarchies model image sort management method based on the marking area rarefaction representation according to claim 1, it is characterized in that: described nearest public parent refers to common parent nearest on the contrary path of two or more a plurality of classes.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2012105048525A CN103246688A (en) | 2012-12-03 | 2012-12-03 | Semantic hierarchy model image classification management method based on salient region sparse representation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2012105048525A CN103246688A (en) | 2012-12-03 | 2012-12-03 | Semantic hierarchy model image classification management method based on salient region sparse representation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103246688A true CN103246688A (en) | 2013-08-14 |
Family
ID=48926212
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2012105048525A Pending CN103246688A (en) | 2012-12-03 | 2012-12-03 | Semantic hierarchy model image classification management method based on salient region sparse representation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103246688A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104636761A (en) * | 2015-03-12 | 2015-05-20 | 华东理工大学 | Image semantic annotation method based on hierarchical segmentation |
| CN105045907A (en) * | 2015-08-10 | 2015-11-11 | 北京工业大学 | Method for constructing visual attention-label-user interest tree for personalized social image recommendation |
| CN108182443A (en) * | 2016-12-08 | 2018-06-19 | 广东精点数据科技股份有限公司 | A kind of image automatic annotation method and device based on decision tree |
| CN114494711A (en) * | 2022-02-25 | 2022-05-13 | 南京星环智能科技有限公司 | Image feature extraction method, device, equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030195883A1 (en) * | 2002-04-15 | 2003-10-16 | International Business Machines Corporation | System and method for measuring image similarity based on semantic meaning |
| CN1936892A (en) * | 2006-10-17 | 2007-03-28 | 浙江大学 | Image content semanteme marking method |
| US20100169318A1 (en) * | 2008-12-30 | 2010-07-01 | Microsoft Corporation | Contextual representations from data streams |
| CN102057371A (en) * | 2008-06-06 | 2011-05-11 | 汤姆逊许可证公司 | System and method for similarity search of images |
| CN102637199A (en) * | 2012-02-29 | 2012-08-15 | 浙江大学 | Image marking method based on semi-supervised subject modeling |
-
2012
- 2012-12-03 CN CN2012105048525A patent/CN103246688A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030195883A1 (en) * | 2002-04-15 | 2003-10-16 | International Business Machines Corporation | System and method for measuring image similarity based on semantic meaning |
| CN1936892A (en) * | 2006-10-17 | 2007-03-28 | 浙江大学 | Image content semanteme marking method |
| CN102057371A (en) * | 2008-06-06 | 2011-05-11 | 汤姆逊许可证公司 | System and method for similarity search of images |
| US20100169318A1 (en) * | 2008-12-30 | 2010-07-01 | Microsoft Corporation | Contextual representations from data streams |
| CN102637199A (en) * | 2012-02-29 | 2012-08-15 | 浙江大学 | Image marking method based on semi-supervised subject modeling |
Non-Patent Citations (1)
| Title |
|---|
| 郑阳: "基于显著区域检测的图像语义层次管理", 《万方数据知识服务平台》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104636761A (en) * | 2015-03-12 | 2015-05-20 | 华东理工大学 | Image semantic annotation method based on hierarchical segmentation |
| CN105045907A (en) * | 2015-08-10 | 2015-11-11 | 北京工业大学 | Method for constructing visual attention-label-user interest tree for personalized social image recommendation |
| CN105045907B (en) * | 2015-08-10 | 2018-03-09 | 北京工业大学 | A kind of construction method of vision attention tagging user interest tree for Personalized society image recommendation |
| CN108182443A (en) * | 2016-12-08 | 2018-06-19 | 广东精点数据科技股份有限公司 | A kind of image automatic annotation method and device based on decision tree |
| CN108182443B (en) * | 2016-12-08 | 2020-08-07 | 广东精点数据科技股份有限公司 | Automatic image labeling method and device based on decision tree |
| CN114494711A (en) * | 2022-02-25 | 2022-05-13 | 南京星环智能科技有限公司 | Image feature extraction method, device, equipment and storage medium |
| CN114494711B (en) * | 2022-02-25 | 2023-10-31 | 南京星环智能科技有限公司 | Image feature extraction method, device, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Weinstein | A computer vision for animal ecology | |
| Cai et al. | A modified YOLOv3 model for fish detection based on MobileNetv1 as backbone | |
| Zheng et al. | Growing status observation for oil palm trees using Unmanned Aerial Vehicle (UAV) images | |
| Zhou et al. | Object detectors emerge in deep scene cnns | |
| Maheswari et al. | Intelligent fruit yield estimation for orchards using deep learning based semantic segmentation techniques—a review | |
| Sivic et al. | Unsupervised discovery of visual object class hierarchies | |
| Zhang et al. | Integrating bottom-up classification and top-down feedback for improving urban land-cover and functional-zone mapping | |
| Liu et al. | Classification of tree species and stock volume estimation in ground forest images using Deep Learning | |
| CN109002834B (en) | Fine-grained image classification method based on multi-modal representation | |
| Wang et al. | Remote sensing image retrieval by scene semantic matching | |
| Wang et al. | YOLOv3‐Litchi Detection Method of Densely Distributed Litchi in Large Vision Scenes | |
| Li et al. | DeepCotton: in-field cotton segmentation using deep fully convolutional network | |
| Rong et al. | Pest Identification and Counting of Yellow Plate in Field Based on Improved Mask R‐CNN | |
| Jia et al. | Accurate segmentation of green fruit based on optimized mask RCNN application in complex orchard | |
| CN103246688A (en) | Semantic hierarchy model image classification management method based on salient region sparse representation | |
| CN103530405A (en) | Image retrieval method based on layered structure | |
| Jenrette et al. | Shark detection and classification with machine learning | |
| CN112613548A (en) | User customized target detection method, system and storage medium based on weak supervised learning | |
| Bouchakwa et al. | A review on visual content-based and users’ tags-based image annotation: methods and techniques | |
| CN102945372B (en) | Classifying method based on multi-label constraint support vector machine | |
| Su et al. | Graph learning on K nearest neighbours for automatic image annotation | |
| Wang et al. | Multimodal Poisson gamma belief network | |
| CN117611988A (en) | Automatic identification and monitoring method and system for newly-increased farmland management and protection attribute | |
| Dimitrovski et al. | Detection of visual concepts and annotation of images using ensembles of trees for hierarchical multi-label classification | |
| Chum et al. | Web scale image clustering |
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: 20130814 |