CN102073748B - Visual keyword based remote sensing image semantic searching method - Google Patents
Visual keyword based remote sensing image semantic searching method Download PDFInfo
- Publication number
- CN102073748B CN102073748B CN2011100546248A CN201110054624A CN102073748B CN 102073748 B CN102073748 B CN 102073748B CN 2011100546248 A CN2011100546248 A CN 2011100546248A CN 201110054624 A CN201110054624 A CN 201110054624A CN 102073748 B CN102073748 B CN 102073748B
- Authority
- CN
- China
- Prior art keywords
- image
- keyword
- feature
- vision
- notable
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Images
Landscapes
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to a visual keyword based remote sensing image semanteme searching method. The method comprises the following steps: setting visual keywords which describe image contents in an image base; selecting a training image from the image base; extracting remarkable visual characteristics of each training image, wherein the remarkable visual characteristics include remarkable points, main dominant tone and texture; acquiring a key mode through a cluster center of a cluster algorithm; establishing a visual keyword hierarchical model by adopting a Gaussian mixture model; extracting the remarkable visual characteristics of all images in the image base, setting weight parameters, and constructing a visual keyword characteristic vector describing the image semanteme; and calculating the similarity between an image to be searched and all images according to the similarity criterion, and outputting a search result according to the high-low sequence of the similarity. The method can effectively improve the recall ratio and the precision ratio of image searching by establishing the correlation between low-layer remarkable visual characteristics and high-layer semantic information through the visual keywords, and the technical scheme provided by the invention has excellent expansibility.
Description
Technical field
The present invention relates to technical field of image processing, more specifically to a kind of remote sensing image semantic retrieving method of view-based access control model keyword.
Background technology
Remote sensing image data application is faced with the contradiction of " data are not only more but also few ".On the one hand, developing rapidly with Aero-Space and various kinds of sensors technology, computer networking technology, database technology etc., particularly retrievable various remote sensing image data products, high spatial resolution remote sense image data are all increasing with surprising rapidity daily；On the other hand, in such immense remote sensing image data warehouse, people, which but generally feel, to be wanted quickly to find target not a duck soup interested.This is due to the features such as remote sensing image data has spatiality, diversity, complexity and magnanimity in itself so that lacks effective search method to mass remote sensing image data at present, has hindered the application of remote sensing image data.The efficient retrieval of remote sensing image is the key for solving the contradiction between mass remote sensing data and people's demand growing to remote sensing data application, is current remote sensing application field problem urgently to be resolved hurrily, is also the forward position of disciplinary study.
Studied in Remote Sensing Image Retrieval in involved every key technology, current research emphasis is concentrated mainly on the visualization feature of remote sensing image（Including spectral signature, textural characteristics, shape facility and assemblage characteristic）In extraction and its similitude matching algorithm, research wherein to textural characteristics and it is most widely used and gos deep into, the problem of being one extremely complex for the description of target shape feature and extraction relative spectral feature, textural characteristics, so far there is no the definite mathematical definition of " shape " also, it is currently based in the video search of content, the shape of target is generally described using edge and provincial characteristics, but for target edge and provincial characteristics describe operator and its shape similarity matching research it is still perfect not to the utmost；The difficulty for being described and being extracted based on target shape feature, although it is increasingly recognised that its significance in Remote Sensing Image Retrieval, achievement in research is extremely limited.In terms of the Remote Sensing Image Retrieval based on assemblage characteristic, mainly there is the retrieval based on tone and texture-combined feature.Data prediction（Automatic Partitioning tissue or pretreatment）It is also based on textural characteristics with the algorithm that visualization feature is combined.
Because low layer visualization feature can not intuitively reflect the semantic information of image, the retrieval result of " required non-gained " under the auxiliary without experts database or domain knowledge base, generally can be all produced.This problem is solved, recall precision and retrieval rate is improved, the dependence to visualization feature must be broken through in search method.Remote sensing image high-level semantics features contain understanding of the people to presentation content, the visual signatures such as color, texture, shape are not only taken into account based on semantic search method, and emphasis is to the semantic description of presentation content, therefore semantic retrieval is more more abundant than the content retrieval of view-based access control model feature, accurate, intelligent higher.However, the Remote Sensing Image Retrieval for being currently based on semanteme remains in the exploratory stage.
The content of the invention
It is an object of the invention in view of the shortcomings of the prior art and not enough, a kind of remote sensing image semantic retrieving method of view-based access control model keyword is provided, by the image analysis methods for meeting human visual perception characteristic, the characteristics of remote sensing image of complexity is abstract for the vision keyword with semantic information, the association set up by vision keyword between low-level image feature, middle level object and high-layer semantic information there is provided method can be suitably used for various types of Remote Sensing Image Retrieval fields.
The technical solution adopted in the present invention is a kind of Remote Sensing Image Retrieval semantic method of view-based access control model keyword, is comprised the following steps：
Step one, the vision keyword of image modality in Image Database can be described by setting, and is selected respectively from Image Database and can be reflected some width images of each vision keyword, be used as and trained image；
Step 2, extracts all kinds of notable visual signatures of all training images；
Obtained all kinds of notable visual signatures, to all training images, are respectively adopted clustering algorithm and clustered, obtained the cluster centre equal with vision keyword number, each cluster centre is mapped as into a critical mode by step 3；The probability density function that any notable visual signature belongs to every class vision keyword is fitted using gauss hybrid models, gauss hybrid models parameter Estimation carrys out self-training image, approximating method uses expectation maximization method of estimation, so as to set up vision key word hierarchy model；
Step 4, extracts all kinds of notable visual signatures of all images in Image Database respectively by the way of consistent with step 2；
Step 5, for each width image in Image Database, the probability that notable visual signature belongs to every class vision keyword is calculated by probability density function obtained by step 3, if belonging to the maximum probability of certain class vision keyword, then think that notable visual signature belongs to such vision keyword, so as to realize notable visual signature to the mapping of vision keyword；
Step 6, for each width image in Image Database, the frequency occurred according to default setting weight parameter, statistics per class vision keyword in the image, and then build the vision keyword feature vector for describing image semanteme；
Step 7, using default similarity measurement criterion, the similitude of image to be retrieved and all images in Image Database is calculated by vision keyword feature vector, retrieval result is sorted and exported from high to low according to similitude.
Moreover, in step 2 and step 4, the notable visual signature of extraction includes dominant hue and texture that significant point, object drive.
Moreover, the implementation for extracting notable visual signature is as follows,
(1) significant point of all training images of operator extraction is described using SIFT image local features, so as to obtain the notable point feature of image, each significant point is represented with 128 dimensional feature vectors；
(2) over-segmentation based on Quick Shift algorithms is carried out to all training images, region merging technique is carried out to over-segmentation result, then HSV models are used to uniformity subject area, the dominant hue in each region is extracted according to the quantized result of its tone passage, so as to obtain the dominant hue feature of image, the dominant hue feature of each subject area is represented with a characteristic vector；
(3) over-segmentation based on Quick Shift algorithms is carried out to all training images, region merging technique is carried out to over-segmentation result, then wavelet transformation is used to uniformity subject area, the average and variance of each yardstick high fdrequency component are obtained as texture descriptor, so as to obtain the textural characteristics of image, the textural characteristics of each subject area are represented with a characteristic vector.
Moreover, in step 6, when setting weight parameter, notable point feature is assigned to average weight, and dominant hue feature and textural characteristics are using the area of itself subject area as weight.
Moreover, in step 3, the clustering algorithm used is K averages or ISODATA algorithm.
Moreover, in step 7, default similarity measurement criterion is the first approximation distance of KL divergences.
Technical scheme that the present invention is provided has the beneficial effect that, associating between the notable visual signature of low layer and high-layer semantic information is set up by the hierarchical model of vision keyword, reduce " semantic gap " between the notable visual signature of low layer and high-level semantic, it is that the quick positioning from mass remote sensing image storehouse and lookup interesting target provide a new solution route, the recall ratio and precision ratio of video search can be effectively improved.The technical scheme that the present invention is provided simultaneously has good autgmentability, significant point, dominant hue and texture that the notable visual signature used includes but is not limited to used in the present invention, as long as meeting the feature of human visual system, can successfully it include in the technical scheme that the present invention is provided.
Brief description of the drawings
Fig. 1 is the flow chart of the embodiment of the present invention.
Fig. 2 is the effect diagram of the embodiment of the present invention.
Embodiment
The remote sensing image semantic retrieving method of view-based access control model keyword proposed by the present invention first sets the vision keyword of reflection Image Database content, selection training image, extract notable visual signature, then vision key word hierarchy model is set up, realize associating between low-level visual feature and high-level semantic, semantic modeling and description are carried out to remote sensing image, finally the image in Image Database retrieved using similarity criteria.It is wherein main to include the notable Visual Feature Retrieval Process of training image, set up vision key word hierarchy model, remote sensing image semantic modeling and the video search Four processes based on similarity criteria.
To describe embodiment in detail, referring to Fig. 1, embodiment flow is as follows：
Step S01, sets the vision keyword for describing presentation content in Image Database.
The data that embodiment is used come from the Zhengzhou area WorldView images of 2009-12-27 collections, image spatial resolution is 0.5 meter, size is 8740*11644, and image is divided into the sub-block of 320*320 sizes according to Tiles partitioned modes, constitutes the retrieval Image Database of 1036 width sub-images.Because remote sensing image area coverage is big, ground species are complicated, and the cover type of earth's surface can be divided into following eight class by the feature showed according to atural object on image：Farmland, open ground, road, compact settlement, sparse population area, square, viaduct, greenery patches.Therefore, embodiment sets eight class vision keywords：Farmland, open ground, road, compact settlement, sparse population area, square, viaduct, greenery patches.
Step S02, according to the vision keyword of setting, the image that can reflect that the atural object content of these keywords is single is found out from Image Database, is used as training image.
Embodiment selects the image blocks pure with farmland, open ground, road, compact settlement, sparse population area, square, viaduct, greenery patches respective type as training sample respectively.
Step S03, using feature extraction algorithm, image is trained to each width, notable visual signature is extracted.
Remote sensing image covering ground species are various, and single feature space is hardly formed effective differentiation to atural object, and embodiment of the present invention selection represents the significant point of local feature, the dominant hue and the notable visual signature of textural characteristics three major types of object driving.During specific implementation, the acceptable feature such as selected shape as needed.
For the sake of ease of implementation, the extraction below to this notable visual signature of three classes is described respectively：
（1）Significant point feature extraction：For remote sensing image, angle point is the key character for representing and analyzing image, the characteristics of image extracted from notable neighborhood of a point can effectively reflect the local message of image, and when people pay close attention to a width image, tend to be attracted by significant part in image, which part visual focus is the angle point in image.The present invention describes the notable point feature of operator extraction using SIFT image local features.SIFT feature vector keeps constant to rotation, scaling, brightness change, and the influence of different spatial resolutions, different illumination conditions to significant point feature extraction can be reduced as far as possible.
（2）Dominant hue feature extraction：Using Quick Shift partitioning algorithms, utilization space uniformity and colour consistency carry out over-segmentation to image, then the subject area obtained by over-segmentation are merged, the subject area after being merged,
HSV models correspond directly to the three elements of human eye color vision feature, and three Color Channels are each independent, and the dominant hue of image is can extract out according to the quantized result of its tone passage.Tone passage is quantified as by the present invention first
Individual sub-district, each subject area extracted after above-mentioned over-segmentation merging is represented with the dominant hue histogram after quantifying respectively,
For a certain region after Image Segmentation, then object
Dominant hue characteristic vector can be expressed as follows shown in formula, whereinFor
The frequency that class tone occurs,
Take 1 to
。
（3）Texture feature extraction：Over-segmentation is carried out to image using Quick Shift partitioning algorithms, then the merging of subject area is carried out using the subject area merging method described in dominant hue feature extracting method, finally describing method using multiple dimensioned multi-direction textural characteristics carries out texture feature extraction.The present invention is used as texture descriptor using the average and variance of each yardstick high fdrequency component after wavelet transformation, feature vector dimension obtained by this description method is low, efficiency high and with certain representativeness, the present invention has carried out normalized to wavelet conversion coefficient simultaneously
For a certain region after Image Segmentation, object
Textural characteristics vector representation be shown below,
For
The normalization average value and variance of individual component,
Take 1 to
,
It is total for component, equal to three times of yardstick quantity.
During specific implementation, it can realize（2）With（3）When, it is over-segmentation process setting qualifications, to improve segmentation quality, for example, limits the region after these merging and meet three below condition：
A. subject area internal differences should be as small as possible；
B. object and difference should be larger between contiguous object around it；
C. subject area area should be greater than a certain threshold value.
Condition a purpose is to limit object as pure end member, improves the accuracy of semantic assignment；The degree that condition b control object regions merge；Condition c main purpose is to reject the trifling region that interference vision judges, prominent significant principal character improves the efficiency of algorithm.Assuming that
For a certain region after Image Segmentation, region
Internal differences
It is defined as：
Wherein
,
Weight is characterized, is met
；
For the standard deviation of region internal color,
For region shape index, it is defined as follows shown：
Obtained all kinds of notable visual signatures, to all training images, are respectively adopted clustering algorithm and clustered, obtained the cluster centre equal with vision keyword number, each cluster centre is mapped as into a critical mode by step S04；The probability density function that any notable visual signature belongs to every class vision keyword is fitted using gauss hybrid models, gauss hybrid models parameter Estimation carrys out self-training image, approximating method uses expectation maximization method of estimation, so as to set up vision key word hierarchy model.
What embodiment set up vision key word hierarchy model implements process：The notable visual signature of three classes that image can be extracted, including significant point, dominant hue and texture are trained by a width.For all characteristic vectors of the notable visual signature of each class, clustered using K averages or ISODATA clustering methods, obtain the cluster centre consistent with vision keyword number, each cluster centre is mapped as a critical mode.The probability density function that any feature vector belongs to every class vision keyword is fitted using gauss hybrid models, model parameter estimation carrys out self-training image, and method uses expectation maximization method of estimation, so as to set up vision key word hierarchy model.Since in embodiment, the generalized Gaussian distribution component included in each classification GMM model（GGD）Quantity is 8.Center namely critical mode by the subspace in training sample image feature space, obtain the Gaussian Profile of each critical mode, independently combinable namely Gaussian Profile the merging of multiple critical modes constitutes one containing semantic keyword, view picture image is expressed as the distribution histogram of all kinds of semantic key words in image, so far can complete without semantic visual signature to the modeling process containing semantic keyword label.
For the sake of ease of implementation, this step offer related description is as follows：
Remote sensing image is represented by from pixel to local notable feature or the hierarchical model of primitive, destination object and scene, a series of visual vocabulary of description visual informations is all included on each level of model, so as to form shadow table up to the connection of semantic label and characteristics of image in scene.If a certain video vision vocabulary definitions are set
, wherein
Identified for lexical types,
For any visual vocabulary element,
It is total for lexical types,
For whole possible lexical space set.
The polymerization of visual vocabulary can produce the image of any yardstick, some of them polymerization belong to global polymerization, these word combinations it is reducible go out image in most information, text in these word combination patterns are referred to as critical mode, based on above formula, a certain critical mode
It is defined as：
Critical mode set need to meet approximate condition for completeness i.e.
, wherein
It is total for critical mode,
For image feature space.Therefore, vision keyword models are the modeling to critical mode.Automatic cluster algorithm is widely used in terms of feature space Automatic-searching cluster centre, and the present invention finds critical mode by conventional K averages or ISODATA automatic clusters algorithm from numerous and jumbled visual vocabulary table；Gauss hybrid models can describe the data distribution in sample space with the method for parametrization, the parameter of gauss hybrid models is had as the feature of image and is concisely and efficiently advantage, assuming that visual vocabulary is distributed Gaussian distributed in feature space, Gaussian Mixture distribution GMMs is obeyed in then critical mode distribution, the semantic key words of each classification i.e. by
The cluster centre composition of individual lexical space, with feature distribution
Exemplified by, it belongs to keywordProbability density function be represented by：
WhereinIt is characterized distribution space,
For the dimension of GMM model,
For vision keyword
GMM model parameter,,
For in modelThe mixed coefficint of individual gaussian variable,
For
The average of individual gaussian variable,
The transposition of representing matrix,
For corresponding covariance matrix,
For sample dimension.Model parameter estimation is using selected image of all categories as training data, and method uses expectation maximization method of estimation, the unique GMM distributions in vision keyword character pair space of each classification.
Below by taking significant point feature as an example, it is described in detail and is fitted the process that any feature vector belongs to the probability density function of every class vision keyword using gauss hybrid models：Use first clustering algorithm by 128 dimension SIFT feature vector clusters for
Sub-spaces, one critical mode of center representative per sub-spaces, it is assumed that each SIFT feature critical mode Gaussian distributed
, then it is fitted using GMM model
The semantic key words of one SIFT type of combinational expression of individual SIFT feature critical mode.
Assuming that image semantic key words number is, then image after training
SIFT type vision key word hierarchy models
It is expressed as shown in following formula：
Wherein
For
The corresponding gauss hybrid models coefficient of individual critical mode,
For
Belong to
The probability distribution of individual keyword, is a Gaussian Profile,
For
The corresponding characteristic vector of individual critical mode,
TheThe covariance matrix of sub-spaces.Calculate extracted significant point descriptionBelong to
The posterior probability of individual keyword
, SIFT is described by vector using maximum a posteriori probability grader MAP (maximum a posteriori)
It is labeled as
Keyword, wherein：
If by every width image regard as by some crucial phrases in vision keywords database into " text ", classical statistical method TF-IDF (term frequent-Inverse document frequency) then in text retrieval technique is the item frequency-row's of falling text frequency, can be used to assess significance level of the words for certain part of file in a file set or a corpus.The number of times that the importance of words occurs hereof with it is directly proportional increase, but the frequency that can occur simultaneously with it in corpus is inversely proportional decline.The present invention is using a single point as statistic unit, image
The SIFT feature vector of view-based access control model keyword can be expressed as follows shown in formula：
The fit procedure of dominant hue and texture is similar with the process of significant point, and it will not go into details by the present invention.Through over-fitting, that is, vision key word hierarchy model is established, realize low-level visual feature to the contact between high-level semantic keyword.
Step S05, using feature extraction algorithm, to each width image in Image Database, extracts notable visual signature.The notable visual signature that embodiment is extracted includes significant point, dominant hue and texture, obtains a series of characteristic vector of description image features, i.e., corresponding significant point characteristic vector, dominant hue characteristic vector and texture feature vector.Specific implementation process is consistent with step S03.
Step S06, according to the vision key word hierarchy model set up in step S04, the characteristic vector of three class visual signatures of each width image in Image Database is substituted into vision key word hierarchy model, the probability for belonging to vision keyword per all characteristic vectors in class visual signature is calculated, presses
Principle by maps feature vectors be corresponding vision keyword, so as to set up the characteristic vector of all visual signatures and the corresponding relation of vision keyword.
Step S07, for each width image in Image Database, according to default setting weight parameter, the frequency that statistics occurs per class vision keyword in the image, and then the vision keyword feature vector for describing image semanteme is built, so as to realize the semantic modeling to image view-based access control model keyword and description.During specific implementation, using different weight parameters, obtained retrieval result is different, can rule of thumb be set in advance by those skilled in the art.
Remote sensing image semantic modeling implements process in total embodiment：In step S05, to all images in Image Database according to above step S03, notable visual signature, including significant point, dominant hue and texture are first extracted；Then in step S06, under the support of vision key word hierarchy model, according to
Principle, the single feature vector of every width Extraction of Image can be mapped as the keyword of finite number,
,
For characteristic point in image or the number of object.But the contribution rate of each keyword identification image is not duplicate, by taking section object as an example, in general the contribution of image center region or larger area region to image interpretation is greater than the less region of corner area, the embodiment of the present invention is in step S07, uniform weight is used to characteristic point, the weight all same of i.e. each point feature critical mode, to region keyword（Dominant hue, texture）Using area factor as weight parameter, then the frequency that each keyword occurs in image, by taking single category feature as an example, image can be counted
Keyword modeling after characteristic vectorIt is represented by
For keyword number,
For keyword
The frequency of appearance,
Take 1 to
, for characteristic point,
,For
The number of times that individual point feature key model occurs, n is the total degree that all point feature key models occur；For dominant hue or textural characteristics,
,
For the image gross area,
For sizes of the keyword i in image.Thus it can obtain the normalization vision keyword feature vector of image.
Step S08, sets the weight of three class vision keyword features vector, and with the first approximation distance of KL divergences for similarity measurement mode, the similitude of image to be retrieved and all images in Image Database is calculated by vision keyword feature vector.During Concrete facts, it would however also be possible to employ other prior arts are used as similarity measurement criterion, such as COS distance, KL distances, the first approximation distance of KL divergences, Euclidean distance, mahalanobis distance.
The video search based on similarity criteria implements process in embodiment：Using feature extraction algorithm, the keyword that three types in image can be obtained describes the significant point, dominant hue and textural characteristics of image respectively, the TF-IDF that the description of keyword is used for reference in document representation method describes method, therefore, each image may be described as the characteristic vector of a certain class or a few class vision keywords
。
Analyzed from the angle of information theory, the probability distribution relation that vision keyword occurs in a width image is meant that expressed by the characteristic vector extracted, it is assumed that separate between each vision keyword, and obey probability density function
Distribution, then the distance between two width images be also referred to as Kullback-Leibler divergences, be shown below：
Wherein subscript
WithDifference represents two width images.
Between vision keyword feature vector
Divergence computing formula is：
Wherein subscript
1 He2 differences represent two width images,It is total for keyword,
For keyword sequence number, above formula includes logarithm operation, and computational efficiency is low, selects its single order approximate distance
Complexity can be effectively reduced, it is as follows：
One width image can be expressed using significant point, dominant hue, the category feature vector of texture three, and the similitude size of image semanteme distribution is calculated using the Weighted distance of three category features, as follows：
Wherein
,
Significant point, dominant hue, texture are represented respectively,
The width image of distinctive mark two.Keyword feature vector expression mode is similar to image histogram in the present invention, simply what it was inputted is the semantic key words of different characteristic type, rather than the gray value of image, this expression way is unrelated with image size, the semantic dimension of setting identical is only needed to, different size of image all can carry out similarity measurement using above formula.
Step S09, similitude is ranked up according to order from high to low, exports retrieval result.
In the semantic key words characteristic vector and Image Database that above step S08 calculates retrieval image after the similitude of the semantic key words characteristic vector of all images, according to the Sequential output result image of similitude from high to low, as retrieval result.
Technical solution of the present invention effect for convenience of description, using the average precision of 8 class images of different weights, single textural characteristics, single dominant hue feature, the situation of comprehensive characteristics used by single notable point feature and the embodiment of the present invention is contrasted to be retrieved respectively, then comparative evaluation is carried out to acquired results, as shown in Figure 2.Quantitative evaluation takes average precision, that is, the ratio containing similar view in the preceding 16 width image returned.Similarity measurement criterion employs the first approximation distance of KL divergences, and significant point, dominant hue, the weight of the class vision keyword of main texture three are respectively set as 0.5,0.25,0.25 in comprehensive characteristics retrieval.It can be seen that, the accuracy rate of technical solution of the present invention is higher.
Above content is to combine optimum implementation to say the further description done to the present invention, it is impossible to assert that the specific implementation of the present invention is only limited to these explanations.It should be appreciated by those skilled in the art in the case where not departing from and being defined by the appended claims, various modifications can be carried out in detail, protection scope of the present invention should be all considered as belonging to.
Claims (4)
1. a kind of remote sensing image semantic retrieving method of view-based access control model keyword, it is characterised in that comprise the following steps：
Step one, the vision keyword of image modality in Image Database can be described by setting, and is selected respectively from Image Database and can be reflected some width images of each vision keyword, be used as and trained image；
Step 2, extracts all kinds of notable visual signatures of all training images, and the notable visual signature of extraction includes dominant hue and texture that significant point, object drive；
Obtained all kinds of notable visual signatures, to all training images, are respectively adopted clustering algorithm and clustered, obtained the cluster centre equal with vision keyword number, each cluster centre is mapped as into a critical mode by step 3；The probability density function that any notable visual signature belongs to every class vision keyword is fitted using gauss hybrid models, gauss hybrid models parameter Estimation carrys out self-training image, approximating method uses expectation maximization method of estimation, so as to set up vision key word hierarchy model；
Step 4, extracts all kinds of notable visual signatures of all images in Image Database respectively by the way of consistent with step 2；
Step 5, for each width image in Image Database, the probability that notable visual signature belongs to every class vision keyword is calculated by probability density function obtained by step 3, if belonging to the maximum probability of certain class vision keyword, then think that notable visual signature belongs to such vision keyword, so as to realize notable visual signature to the mapping of vision keyword；
Step 6, for each width image in Image Database, the frequency occurred according to default setting weight parameter, statistics per class vision keyword in the image, and then build the vision keyword feature vector for describing image semanteme；Carry out setting weight parameter it is default when, notable point feature is assigned to average weight, and dominant hue feature and textural characteristics are using the area of itself subject area as weight；
Step 7, using default similarity measurement criterion, the similitude of image to be retrieved and all images in Image Database is calculated by vision keyword feature vector, retrieval result is sorted and exported from high to low according to similitude.
2. the remote sensing image semantic retrieving method of view-based access control model keyword according to claim 1, it is characterised in that：The implementation for extracting notable visual signature is as follows,
(1) significant point of all training images of operator extraction is described using SIFT image local features, so as to obtain the notable point feature of image, each significant point is represented with 128 dimensional feature vectors；
(2) over-segmentation based on Quick Shift algorithms is carried out to all training images, region merging technique is carried out to over-segmentation result, then HSV models are used to uniformity subject area, the dominant hue in each region is extracted according to the quantized result of its tone passage, so as to obtain the dominant hue feature of image, the dominant hue feature of each subject area is represented with a characteristic vector；
(3) over-segmentation based on Quick Shift algorithms is carried out to all training images, region merging technique is carried out to over-segmentation result, then wavelet transformation is used to uniformity subject area, the average and variance of each yardstick high fdrequency component are obtained as texture descriptor, so as to obtain the textural characteristics of image, the textural characteristics of each subject area are represented with a characteristic vector.
3. the remote sensing image semantic retrieving method of view-based access control model keyword according to claim 1 or claim 2, it is characterised in that：In step 3, the clustering algorithm used is K averages or ISODATA algorithm.
4. the remote sensing image semantic retrieving method of view-based access control model keyword according to claim 1 or claim 2, it is characterised in that：In step 7, default similarity measurement criterion is the first approximation distance of KL divergences.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011100546248A CN102073748B (en) | 2011-03-08 | 2011-03-08 | Visual keyword based remote sensing image semantic searching method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011100546248A CN102073748B (en) | 2011-03-08 | 2011-03-08 | Visual keyword based remote sensing image semantic searching method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102073748A CN102073748A (en) | 2011-05-25 |
| CN102073748B true CN102073748B (en) | 2012-07-25 |
Family
ID=44032287
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2011100546248A Expired - Fee Related CN102073748B (en) | 2011-03-08 | 2011-03-08 | Visual keyword based remote sensing image semantic searching method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102073748B (en) |
Families Citing this family (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103020172B (en) * | 2012-11-28 | 2015-08-19 | 北京京东世纪贸易有限公司 | A kind of method and apparatus utilizing video information to search for article |
| CN103177105B (en) * | 2013-03-26 | 2016-12-28 | 新浪网技术(中国)有限公司 | A kind of image search method and device |
| CN103823843B (en) * | 2014-01-24 | 2017-02-22 | 北京理工大学 | Gauss mixture model tree and incremental clustering method thereof |
| CN104834757A (en) * | 2015-06-05 | 2015-08-12 | 昆山国显光电有限公司 | Image semantic retrieval method and system |
| CN106021250A (en) * | 2015-09-16 | 2016-10-12 | 展视网（北京）科技有限公司 | Image semantic information retrieval method based on keyword |
| CN105224619B (en) * | 2015-09-18 | 2018-06-05 | 中国科学院计算技术研究所 | A kind of spatial relationship matching process and system suitable for video/image local feature |
| CN105243154B (en) * | 2015-10-27 | 2018-08-21 | 武汉大学 | Remote sensing image retrieval method based on notable point feature and sparse own coding and system |
| CN105608673B (en) * | 2015-12-16 | 2020-09-25 | 清华大学 | Image color quantization and dithering method and system |
| CN106570136B (en) * | 2016-11-02 | 2019-10-29 | 中国科学院深圳先进技术研究院 | A kind of remote sensing images semantic retrieving method and device based on Pixel-level correlation rule |
| CN106570125B (en) * | 2016-11-02 | 2020-08-25 | 中国科学院深圳先进技术研究院 | Remote sensing image retrieval method and device for rotational scaling and translation invariance |
| CN106777094A (en) * | 2016-12-14 | 2017-05-31 | 大连大学 | The medical science big data searching system of the Skyline that view-based access control model vocabulary is matched with multiple features |
| CN106777090A (en) * | 2016-12-14 | 2017-05-31 | 大连交通大学 | The medical science big data search method of the Skyline that view-based access control model vocabulary is matched with multiple features |
| CN107862238B (en) * | 2016-12-26 | 2021-05-14 | 北京理工雷科电子信息技术有限公司 | On-orbit aircraft candidate area screening method based on local texture density and divergence |
| CN108629345B (en) | 2017-03-17 | 2021-07-30 | 北京京东尚科信息技术有限公司 | High-dimensional image feature matching method and device |
| CN107291855A (en) * | 2017-06-09 | 2017-10-24 | 中国电子科技集团公司第五十四研究所 | A kind of image search method and system based on notable object |
| CN107423695A (en) * | 2017-07-13 | 2017-12-01 | 苏州珂锐铁电气科技有限公司 | Dynamic texture identification method based on bipartite graph |
| CN107909561B (en) * | 2017-11-15 | 2021-07-27 | 浙江大学宁波理工学院 | Optimal transmission image color migration method based on weight optimization |
| CN108230413B (en) * | 2018-01-23 | 2021-07-06 | 北京市商汤科技开发有限公司 | Image description method and device, electronic equipment and computer storage medium |
| CN108446616B (en) * | 2018-03-09 | 2021-09-03 | 西安电子科技大学 | Road extraction method based on full convolution neural network ensemble learning |
| WO2021042763A1 (en) * | 2019-09-03 | 2021-03-11 | Guangdong Oppo Mobile Telecommunications Corp., Ltd. | Image searches based on word vectors and image vectors |
| CN110956213A (en) * | 2019-11-29 | 2020-04-03 | 珠海大横琴科技发展有限公司 | Method and device for generating remote sensing image feature library and method and device for retrieving remote sensing image |
| CN111291319B (en) * | 2020-03-24 | 2023-07-07 | 广东海洋大学深圳研究院 | Mobile robot state estimation method applied to non-Gaussian noise environment |
| CN113139378B (en) * | 2021-03-18 | 2022-02-18 | 杭州电子科技大学 | Image description method based on visual embedding and condition normalization |
| CN112836069B (en) * | 2021-04-22 | 2021-09-10 | 泰德网聚(北京)科技股份有限公司 | Self-recognition retrieval system based on image keywords |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101241504A (en) * | 2008-01-23 | 2008-08-13 | 武汉大学 | Remote sense image data intelligent search method based on content |
| CN101315663A (en) * | 2008-06-25 | 2008-12-03 | 中国人民解放军国防科学技术大学 | Nature scene image classification method based on area dormant semantic characteristic |
| CN101692224A (en) * | 2009-07-08 | 2010-04-07 | 南京师范大学 | High-resolution remote sensing image search method fused with spatial relation semantics |
-
2011
- 2011-03-08 CN CN2011100546248A patent/CN102073748B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101241504A (en) * | 2008-01-23 | 2008-08-13 | 武汉大学 | Remote sense image data intelligent search method based on content |
| CN101315663A (en) * | 2008-06-25 | 2008-12-03 | 中国人民解放军国防科学技术大学 | Nature scene image classification method based on area dormant semantic characteristic |
| CN101692224A (en) * | 2009-07-08 | 2010-04-07 | 南京师范大学 | High-resolution remote sensing image search method fused with spatial relation semantics |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102073748A (en) | 2011-05-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102073748B (en) | Visual keyword based remote sensing image semantic searching method | |
| CN106649487B (en) | Image retrieval method based on interest target | |
| US10102443B1 (en) | Hierarchical conditional random field model for labeling and segmenting images | |
| Fauqueur et al. | Region-based image retrieval: Fast coarse segmentation and fine color description | |
| US10528620B2 (en) | Color sketch image searching | |
| CN108595636A (en) | The image search method of cartographical sketching based on depth cross-module state correlation study | |
| Kim et al. | Color–texture segmentation using unsupervised graph cuts | |
| Srivastava et al. | A review: color feature extraction methods for content based image retrieval | |
| CN108647602B (en) | A kind of aerial remote sensing images scene classification method determined based on image complexity | |
| CN103377376A (en) | Method and system for image classification, and method and system for image retrieval | |
| CN103186538A (en) | Image classification method, image classification device, image retrieval method and image retrieval device | |
| CN109934258B (en) | Image retrieval method based on feature weighting and region integration | |
| Wang et al. | Remote-sensing image retrieval by combining image visual and semantic features | |
| Yan et al. | Adaptive fusion of color and spatial features for noise-robust retrieval of colored logo and trademark images | |
| Jin et al. | Content-based image retrieval based on shape similarity calculation | |
| Wang et al. | Unsupervised segmentation of greenhouse plant images based on modified Latent Dirichlet Allocation | |
| James | Face Image retrieval with HSV color space using clustering techniques | |
| Meng et al. | Merged region based image retrieval | |
| Jaswal et al. | Content based image retrieval using color space approaches | |
| Ji et al. | Research on indoor scene classification mechanism based on multiple descriptors fusion | |
| Jyothi et al. | Computational color naming for human-machine interaction | |
| Devesh et al. | Retrieval of monuments images through ACO optimization approach | |
| Kumar et al. | Automatic feature weight determination using indexing and pseudo-relevance feedback for multi-feature content-based image retrieval | |
| Zhao et al. | Image retrieval based on color features and information entropy | |
| CN110929631A (en) | Scene classification method based on Lie-AdaBoost remote sensing 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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20120725 |