CN102591918A - Remote sensing image retrieval method based on multi-agent system - Google Patents
Remote sensing image retrieval method based on multi-agent system Download PDFInfo
- Publication number
- CN102591918A CN102591918A CN2011104247758A CN201110424775A CN102591918A CN 102591918 A CN102591918 A CN 102591918A CN 2011104247758 A CN2011104247758 A CN 2011104247758A CN 201110424775 A CN201110424775 A CN 201110424775A CN 102591918 A CN102591918 A CN 102591918A
- Authority
- CN
- China
- Prior art keywords
- image
- similarity
- remote sensing
- color characteristic
- textural characteristics
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention relates to a remote sensing image retrieval method based on a multi-agent system, which includes the steps: parallelly extracting various visual features in a remote sensing image and parallelly calculating similarity of the visual features by designing and realizing respective task division and collaboration modes of various agents including a feature agent, a collaboration agent, a coarse detection agent and a precision detection agent; and adaptively optimizing the occupied weight of various visual features in comprehensive features by means of studying of user feedback and mutual collaboration of the multiple agents, and finally realizing remote sensing image retrieval accordant with human visual sensing characteristics. Compared with a traditional method, the remote sensing image retrieval method has the advantages that the shortcomings of simplicity and objectivity of the retrieval features and poor retrieval efficiency in traditional remote sensing image retrieval are effectively improved, so that retrieval performance is effectively improved.
Description
Technical field
The present invention relates to the image processing technique field, more particularly, the present invention relates to a kind of remote sensing image search method based on multi-agent system.
Background technology
Along with the develop rapidly of space exploration technology, sensor technology, network technology, database technology etc., retrievable remote sensing image data every day is all with (exponential) sharp increase with surprising rapidity; But meanwhile, people face the covered situation of a large amount of useful informations again, reason be handle at present and analyze mass data limited in one's ability, still lack video search means efficiently.The efficient retrieval of remote sensing image is to solve mass remote sensing data and people remotely-sensed data is used the key of the contradiction between the growing demand, is the difficult problem that present remote sensing application field needs to be resolved hurrily, and has represented the forward position of subject research.
Present research is mainly described and is retrieved image through all kinds of visual signatures that extract remote sensing image, is absorbed in the research of video search algorithm mostly, through improving the precision that feature extraction and matching algorithm improve video search.But in practical application, often there is the problem of three aspects: the first, because the type of ground objects in the remote sensing image is very complicated, single visual signature is difficult to realize the effective differentiation to atural object; The second, visual signature only can be represented the objective characteristics of image, according to the user satisfaction of result for retrieval is not carried out the study of intelligence; The 3rd, the traditional serial search method has had a strong impact on recall precision.
Summary of the invention
The object of the present invention is to provide a kind of remote sensing image search method based on multi-agent system; It utilizes the co-operating characteristic of intelligent body in the multi-agent system; The parallel all kinds of visual signatures that extract in the remote sensing image; And pass through the study of field feedback and the mutual cooperation between the multiple agent; Optimize all kinds of visual signatures shared weight in comprehensive characteristics adaptively, the final remote sensing image retrieval that realizes meeting the human visual perception characteristic is for the real time implementation that advances retrieving in magnanimity treatment of remote and the service, intellectuality, personalization etc. provide strong technical support.
The present invention realizes through following technical scheme:
A kind of remote sensing image search method based on multi-agent system may further comprise the steps:
Step 1, the feature intelligent body extracts the visual signature of all images in inquiry image and the remote sensing image storehouse;
Step 2, intelligent body all the Rough Inspection intelligence bodies in system of cooperating send retrieval request;
Step 3, Rough Inspection intelligence body is calculated as follows the comprehensive similarity of all images in inquiry image and the remote sensing image storehouse,
Wherein, N is the number of all images in the remote sensing image storehouse, and K is the species number of visual signature, Q and T
iI width of cloth image in image and the remote sensing image storehouse, W are inquired about in expression respectively
IkK class visual signature shared weights in comprehensive similarity of representing i width of cloth image satisfy
Sim
Ik(Q, T
i) similarity between the proper vector of k class visual signature of expression inquiry image and i width of cloth image;
Step 4, Rough Inspection intelligence body carries out descending sort according to comprehensive similarity, and the preliminary search result { T that preceding M comprehensive similarity is corresponding
j}
J=1 ..., MReturn to the intelligent body of cooperation as input;
Step 5, the user carries out mark to the image in the input according to preset quantitative evaluation grade, and the intelligent body of cooperating feeds back to the intelligent body of smart inspection with user's mark result;
Step 6, the intelligent body of smart inspection is adjusted image { T in inquiry image and the input according to user's mark result
j}
J=1 ..., MThe proper vector of all K class visual signatures between the shared weights of similarity, be designated as { W
Jk} '
J=1,2 ..., M; K=1 ..., K
Step 7, the intelligent body of smart inspection recomputates the comprehensive similarity of all images in inquiry image and the remote sensing image storehouse according to following formula;
Step 8, the intelligent body of smart inspection carries out descending sort again according to the comprehensive similarity after calculating, and result for retrieval { T after the renewal of the correspondence of the comprehensive similarity after will calculating
j} '
J=1 ..., MReturn to the intelligent body of cooperation;
Step 9, contrast are upgraded the image number that satisfies height correlation in the image number that satisfies height correlation in the result for retrieval of back and the input, whether judge difference greater than certain predetermined threshold value, if result for retrieval is as input, even { T after then will upgrading
j}
J=(1 ..., M)={ T
j} '
J=(1 ..., M), { W
Jk}
J=(1 ..., M), k=(1 ..., K)={ W
Jk} '
J=(1 ..., M), k=(1 ..., K), and return step 5, if not, the intelligent body of then cooperating will upgrade the back result for retrieval as final result for retrieval output.
Visual signature comprises color characteristic, textural characteristics and shape facility;
The feature intelligent body comprises color characteristic intelligence body, textural characteristics intelligence body and shape facility intelligence body;
Rough Inspection intelligence body comprises color Rough Inspection intelligence body, texture Rough Inspection intelligence body and shape Rough Inspection intelligence body;
Step 1 comprises following substep:
(1) for any width of cloth image I, the intelligent body of color characteristic, textural characteristics intelligence body and parallel color characteristic, textural characteristics and the shape facility that extracts image of shape facility intelligence body, specifically implementation does,
The hue histogram of color characteristic intelligence body extraction image and saturation degree histogram are to describe the color characteristic of image, and the proper vector of color characteristic is expressed as follows;
FC
I=(h
1,s
1,h
2,s
2,...,h
L,s
L)
Wherein, FC
IThe proper vector of expression color characteristic, L representes that the quantized interval of image tone component and saturation degree component in tone-saturation degree color space is big or small, h
lAnd s
lRepresent l tone component and l the probability that the saturation degree component occurs in whole quantification space respectively, and l=(1,2 ..., L);
Textural characteristics intelligence body at first carries out wavelet decomposition to image, adopts the average of each wavelet sub-band and the textural characteristics that standard deviation is described image then, and the proper vector of textural characteristics is expressed as follows;
Wherein, FT
IThe proper vector of expression textural characteristics, L representes high fdrequency component sum that image is carried out obtaining after the wavelet decomposition,
With
Wavelet coefficient normalization average value and the standard deviation of representing l high fdrequency component respectively, and l=(1,2 ..., L);
Shape facility intelligence body at first adopts edge detection operator to extract edge images to image, utilizes invariant moments to describe the shape facility of image then, and the proper vector of shape facility is expressed as follows:
Wherein, FS
IThe proper vector of expression shape facility,
L Hu invariant moments group component of the edge images that expression employing Canny edge detection operator obtains, l=(1,2 ..., 7).
(2) proper vector of color characteristic, textural characteristics and the shape facility of comprehensive image is formed the video vision characteristic according to following formula.
F
I={FC
I,FT
I,FS
I}
Step 3 comprises following substep:
(1) for any two width of cloth image I and J, color characteristic similarity, textural characteristics similarity and the shape facility similarity of color Rough Inspection intelligence body, texture Rough Inspection intelligence body and shape Rough Inspection intelligence body parallel computation two width of cloth images, specifically implementation does,
Color Rough Inspection intelligence body calculates the color characteristic similarity of two width of cloth images according to following formula,
Wherein,
and
representes l frequency that the tone component occurs at whole quantized interval in two width of cloth images respectively;
and
representes the frequency that l saturation degree component of two width of cloth images occurs at whole quantized interval respectively, and L is the quantized interval size; For the value that guarantees the color characteristic similarity between 0 to 1, need carry out Gaussian normalization according to following formula to the color characteristic similarity,
Wherein, μ and σ represent color characteristic similarity average and the standard deviation that normalization is preceding respectively;
Texture Rough Inspection intelligence body calculates the textural characteristics similarity of two width of cloth images according to following formula,
Wherein,
and
representes the normalization average value and the standard deviation of l high fdrequency component after two width of cloth image wavelet decomposition respectively, and L is the high fdrequency component sum that wavelet decomposition obtains; For the value that guarantees the textural characteristics similarity between 0 to 1, need carry out Gaussian normalization according to following formula to the textural characteristics similarity,
Shape Rough Inspection intelligence body calculates the shape facility similarity of two width of cloth images according to following formula,
Wherein, l corresponding Hu invariant moments group component of
and
expression two width of cloth images; For the value that guarantees the shape facility similarity between 0 to 1, need carry out Gaussian normalization according to following formula to the shapes textures characteristic similarity,
(2) the intelligent body of smart inspection is given initial weight W respectively to color characteristic similarity, textural characteristics similarity and shape facility similarity
C, W
TAnd W
S, and calculate the comprehensive similarity of two width of cloth images according to following formula;
Sim(I,J)=W
CSim
C′(I,J)+W
TSim
T′(I,J)+W
SSim
S′(I,J)
Wherein, W
C, W
TAnd W
SSatisfy W
C+ W
T+ W
S=1;
Quantitative evaluation grade preset in the step 5 comprises height correlation, is correlated with, does not select, has nothing to do and highly irrelevant five grades, and corresponding value is as follows,
Wherein, i=(1,2 ..., 5).
Step 6 comprises following substep:
Wherein, W
Jk' and W
JkRespectively the intelligent body of the smart inspection of expression according to user's mark result to color characteristic, textural characteristics and shape facility in comprehensive similarity shared weight after adjustment and the value before the adjustment, { T
j}
J=1 ..., MThe similar image set that the intelligent body of expression cooperation returns, { T
JC}
J=1,2 ..., M, { T
JT}
J=1,2 ..., M{ T
JS}
J=1,2 ..., MRepresent respectively to gather according to the similar image that color characteristic similarity, textural characteristics similarity and shape characteristic similarity obtain separately.As the W that calculates according to following formula
Jk, make W at '<0 o'clock
Jk'=0;
(2) according to following formula weights are carried out normalization,
The present invention has following advantage and technique effect:
(1) the present invention utilizes the characteristic of intelligent body in the multi-agent system; Extract all kinds of visual signatures in the remote sensing image concurrently and calculate all kinds of characteristic similarities; Through intelligent body to the study of field feedback and the mutual cooperation between the multiple agent; Optimize all kinds of visual signatures shared weight in comprehensive characteristics adaptively, can improve the precision ratio and the recall precision thereof of video search effectively.
(2) technical scheme provided by the invention has good extendability; The visual signature that is adopted includes but not limited to employed color, texture and shape among the present invention; So long as meet the characteristic of human visual perception characteristic, can both successfully include in the technical scheme provided by the invention.
Description of drawings
Fig. 1 is the process flow diagram that the present invention is based on the remote sensing image search method of multi-agent system.
Fig. 2 is the refinement process flow diagram of step 1 in the inventive method.
Fig. 3 is the refinement process flow diagram of step 3 in the inventive method.
Fig. 4 is the refinement process flow diagram of step 6 in the inventive method.
Fig. 5 illustrates the first retrieval effectiveness figure based on comprehensive characteristics.
Fig. 6 illustrates and adopts the retrieval effectiveness figure that adjusts through weights in the inventive method.
Fig. 7 illustrates the retrieval effectiveness figure of single visual signature.
Fig. 8 illustrates the retrieval effectiveness figure that adopts in the inventive method based on the comprehensive characteristics of multi-agent system.
Fig. 9 illustrates and adopts the inventive method and do not adopt the performance comparison figure of the inventive method aspect average precision ratio.
Figure 10 illustrates and adopts the inventive method and do not adopt the performance comparison figure of the inventive method aspect efficient.
Embodiment
Below at first technical term of the present invention is made an explanation and explain.
The feature intelligent body is meant the intelligent body of accomplishing image integration Visual Feature Retrieval Process function.
The intelligent body of cooperating is meant the intelligent body of accomplishing mutual collaboration feature between Rough Inspection intelligence body, user and the intelligent body of smart inspection.
Rough Inspection intelligence body is meant the intelligent body of comprehensive similarity computing function between the comprehensive visual signature of accomplishing all images in inquiry image and the remote sensing image storehouse.
The intelligent body of smart inspection is meant that completion adjusts all kinds of visual signatures shared weights and recomputate the intelligent body of comprehensive similarity function in comprehensive visual signature according to the user to preliminary search result's mark.
Below the experimental data of using among the embodiment is explained: this experiment is being example from the Zhengzhou area WorldView image of gathering on Dec 27th, 2009; Image spatial resolution is 0.5 meter; Be of a size of 8740 * 11644 pixels; Image is divided into the sub-piece that is of a size of 256 * 256 pixels according to the Tiles partitioned mode, and selects 6 types of more single images of face of land cover type, be respectively farmland, open ground, road, intensive residential district, sparse residential district and greenery patches; Every type of image 50 width of cloth, totally 300 width of cloth are formed the remote sensing image storehouse.
Below in conjunction with accompanying drawing and above experimental data, the present invention is elaborated.
As shown in Figure 1, the present invention is based on the remote sensing image search method of multi-agent system, may further comprise the steps:
Step 1, the feature intelligent body extracts the visual signature of all images in inquiry image and the remote sensing image storehouse.In the present embodiment, the color characteristic, textural characteristics and the shape facility that extract image are as visual signature.When extracting color characteristic, image tone component and saturation degree component in tone-saturation degree color space all are quantified as 16 intervals, so the proper vector of color characteristic can be expressed as FC
I=(h
1, s
1, h
2, s
2..., h
16, s
16); During texture feature extraction, image is carried out 2 layers of wavelet decomposition, obtain 6 high fdrequency components altogether, so the proper vector of textural characteristics can be expressed as
The proper vector of shape facility is represented with the Hu invariant moments group of corresponding edge image, can be expressed as
Step 2, intelligent body all the Rough Inspection intelligence bodies in system of cooperating send retrieval request.In an embodiment, the Rough Inspection intelligence body of reception retrieval request comprises color Rough Inspection intelligence body, texture Rough Inspection intelligence body, shape Rough Inspection intelligence body and the intelligent body of essence inspection.
Step 3, Rough Inspection intelligence body is calculated as follows the comprehensive similarity of all images in inquiry image and the remote sensing image storehouse,
Wherein, N is the number of all images in the remote sensing image storehouse, and K is the species number of visual signature, Q and T
iI width of cloth image in image and the remote sensing image storehouse, W are inquired about in expression respectively
IkK class visual signature shared weights in comprehensive similarity of representing i width of cloth image satisfy
Sim
Ik(Q, T
i) similarity between the proper vector of k class visual signature of expression inquiry image and i width of cloth image.In the present embodiment, K=3, color, texture and shape totally three types of visual signatures have been adopted in expression; The above three types of visual signatures of
expression shared weights in comprehensive similarity equate.
Step 4, Rough Inspection intelligence body carries out descending sort according to comprehensive similarity, and the preliminary search result { T that preceding M comprehensive similarity is corresponding
j}
J=1 ..., MReturn to the intelligent body of cooperation as input.In the present embodiment, M=50.
Step 5, the user carries out mark to the image in the input according to preset quantitative evaluation grade, and the intelligent body of cooperating feeds back to the intelligent body of smart inspection with user's mark result.In the present embodiment, preset quantitative evaluation grade is divided into height correlation, be correlated with, does not select, has nothing to do, 5 grades highly have nothing to do.
Step 6, the intelligent body of smart inspection is adjusted image { T in inquiry image and the input according to user's mark result
j}
J=1 ..., MThe proper vector of all K class visual signatures between the shared weights of similarity, be designated as { W
Jk} '
J=1,2 ..., M; K=1 ..., K
Step 7, the intelligent body of smart inspection recomputates the comprehensive similarity of all images in inquiry image and the remote sensing image storehouse according to following formula.
Step 8, the intelligent body of smart inspection carries out descending sort again according to the comprehensive similarity after calculating, and result for retrieval { T after the renewal of the correspondence of the comprehensive similarity after will calculating
j} '
J=1 ..., MReturn to the intelligent body of cooperation.
Step 9, contrast are upgraded the image number that satisfies height correlation in the image number that satisfies height correlation in the result for retrieval of back and the input, whether judge difference greater than certain predetermined threshold value, if result for retrieval is as input, even { T after then will upgrading
j}
J=(1 ..., M)={ T
j} '
J=(1 ..., M), { W
Jk}
J=(1 ..., M), k=(1 ..., K)={ W
Jk} '
J=(1 ..., M), k=(1 ..., K), and return step 5, if not, the intelligent body of then cooperating will upgrade the back result for retrieval as final result for retrieval output.In the present embodiment, predetermined threshold value is 1.
As shown in Figure 2, the step 1 in the inventive method comprises following substep:
(1) for any width of cloth image I, the intelligent body of color characteristic, textural characteristics intelligence body and parallel color characteristic, textural characteristics and the shape facility that extracts image of shape facility intelligence body, specifically implementation does,
The hue histogram of color characteristic intelligence body extraction image and saturation degree histogram are to describe the color characteristic of image, and the proper vector of color characteristic is expressed as follows;
FC
I=(h
1,s
1,h
2,s
2,...,h
L,s
L)
Wherein, FC
IThe proper vector of expression color characteristic, L representes that the quantized interval of image tone component and saturation degree component in tone-saturation degree color space is big or small, h
lAnd s
lRepresent l tone component and l the probability that the saturation degree component occurs in whole quantification space respectively, and l=(1,2 ..., L);
Textural characteristics intelligence body at first carries out wavelet decomposition to image, adopts the average of each wavelet sub-band and the textural characteristics that standard deviation is described image then, and the proper vector of textural characteristics is expressed as follows;
Wherein, FT
IThe proper vector of expression textural characteristics, L representes high fdrequency component sum that image is carried out obtaining after the wavelet decomposition,
With
Wavelet coefficient normalization average value and the standard deviation of representing l high fdrequency component respectively, and l=(1,2 ..., L);
Shape facility intelligence body at first adopts edge detection operator to extract edge images to image, utilizes invariant moments to describe the shape facility of image then, and the proper vector of shape facility is expressed as follows:
Wherein, FS
IThe proper vector of expression shape facility,
L Hu invariant moments group component of the edge images that expression employing Canny edge detection operator obtains, l=(1,2 ..., 7).
(2) proper vector of color characteristic, textural characteristics and the shape facility of comprehensive image is formed the video vision characteristic according to following formula.
F
I={FC
I,FT
I,FS
I}
As shown in Figure 3, step 3 of the present invention comprises following substep:
(1) for any two width of cloth image I and J, color characteristic similarity, textural characteristics similarity and the shape facility similarity of color Rough Inspection intelligence body, texture Rough Inspection intelligence body and shape Rough Inspection intelligence body parallel computation two width of cloth images, specifically implementation does,
Color Rough Inspection intelligence body calculates the color characteristic similarity of two width of cloth images according to following formula,
Wherein,
and
representes l frequency that the tone component occurs at whole quantized interval in two width of cloth images respectively;
and
representes the frequency that l saturation degree component of two width of cloth images occurs at whole quantized interval respectively, and L is the quantized interval size; For the value that guarantees the color characteristic similarity between 0 to 1, need carry out Gaussian normalization according to following formula to the color characteristic similarity,
Wherein, μ and σ represent color characteristic similarity average and the standard deviation that normalization is preceding respectively;
Texture Rough Inspection intelligence body calculates the textural characteristics similarity of two width of cloth images according to following formula,
Wherein,
and
representes the normalization average value and the standard deviation of l high fdrequency component after two width of cloth image wavelet decomposition respectively, and L is the high fdrequency component sum that wavelet decomposition obtains; For the value that guarantees the textural characteristics similarity between 0 to 1, need carry out Gaussian normalization according to following formula to the textural characteristics similarity,
Shape Rough Inspection intelligence body calculates the shape facility similarity of two width of cloth images according to following formula,
Wherein, l corresponding Hu invariant moments group component of
and
expression two width of cloth images; For the value that guarantees the shape facility similarity between 0 to 1, need carry out Gaussian normalization according to following formula to the shapes textures characteristic similarity,
(2) the intelligent body of smart inspection is given initial weight W respectively to color characteristic similarity, textural characteristics similarity and shape facility similarity
C, W
TAnd W
S, and calculate the comprehensive similarity of two width of cloth images according to following formula;
Sim(I,J)=W
CSim
C′(I,J)+W
TSim
T′(I,J)+W
SSim
S′(I,J)
Wherein, W
C, W
TAnd W
SSatisfy W
C+ W
T+ W
S=1.
As shown in Figure 4, step 6 of the present invention comprises following substep:
Wherein, W
Jk' and W
JkRespectively the intelligent body of the smart inspection of expression according to user's mark result to color characteristic, textural characteristics and shape facility in comprehensive similarity shared weight after adjustment and the value before the adjustment, { T
j}
J=1 ..., MThe similar image set that the intelligent body of expression cooperation returns, { T
JC}
J=1,2 ..., M, { T
JT}
J=1,2 ..., M{ T
JS}
J=1,2 ..., MRepresent respectively to gather according to the similar image that color characteristic similarity, textural characteristics similarity and shape characteristic similarity obtain separately.As the W that calculates according to following formula
Jk, make W at '<0 o'clock
Jk'=0;
(2) according to following formula weights are carried out normalization,
Below the design sketch among the present invention is elaborated.
Retrieval effectiveness figure when Fig. 5 and Fig. 6 illustrate the atural object of selecting intensive residential district type for use and retrieve as the inquiry image.Fig. 5 is for adopting the first retrieval retrieval effectiveness figure of comprehensive characteristics, and precision ratio is 76% in this retrieval; Fig. 6 is for adopting the retrieval effectiveness figure that adjusts through weights in the inventive method, and precision ratio is 92%, it is thus clear that be significantly improved through weights adjustment precision ratio; Also can find out effective raising of adopting method of the present invention can realize retrieval performance from visual effect.
Retrieval effectiveness figure when Fig. 7 and Fig. 8 illustrate the atural object of selecting the urban green space type for use and retrieve as the inquiry image.Fig. 5 is for adopting the retrieval effectiveness figure (selecting color characteristic here for use) of single visual signature, and precision ratio is 80% in this retrieval; Fig. 7 is for adopting in the inventive method the retrieval effectiveness figure based on the comprehensive characteristics of multi-agent system, and precision ratio is 88% in this retrieval, and the visible comprehensive characteristics precision ratio that adopts is significantly improved; Also can find out effective raising of adopting method of the present invention can realize retrieval performance from visual effect.
For better advantage of the present invention must be described, further the inventive method be carried out comprehensive evaluation below, comprise and adopt the inventive method and do not adopt the inventive method aspect average precision ratio and in the performance comparison aspect the efficient.At first; Inquiry image to six types of different types of ground objects; Retrieve respectively based on the situation based on the comprehensive characteristics of multi-agent system in the comprehensive characteristics of solid color characteristic, single textural characteristics, single shape facility, no user evaluation feedback and the inventive method; Get the ratio that contains similar image in preceding 50 width of cloth images that return and carry out quantitative evaluation as average precision ratio; Comparing result is as shown in Figure 9, can find out, adopts the inventive method can have higher average precision ratio; From the efficient aspect the inventive method is estimated then; Same inquiry image to above six types of types of ground objects; Contrast and traditional do not retrieve respectively based on the situation based on the comprehensive characteristics of multi-agent system in multi-agent system and the inventive method, get the T.T. of feature extraction and retrieval and carry out quantitative evaluation, comparing result is shown in figure 10; Can find out, adopt the inventive method can have higher recall precision.
Claims (6)
1. the remote sensing image search method based on multi-agent system is characterized in that, may further comprise the steps:
Step 1, the feature intelligent body extracts the visual signature of all images in inquiry image and the remote sensing image storehouse;
Step 2, intelligent body all the Rough Inspection intelligence bodies in system of cooperating send retrieval request;
Step 3, said Rough Inspection intelligence body calculates the comprehensive similarity of all images in said inquiry image and the said remote sensing image storehouse by following formula,
Wherein, N is the number of all images in the said remote sensing image storehouse, and K is the species number of said visual signature, Q and T
iRepresent i width of cloth image in said inquiry image and the said remote sensing image storehouse respectively, W
IkK class visual signature shared weights in said comprehensive similarity of representing said i width of cloth image satisfy
Sim
Ik(Q, T
i) similarity between the proper vector of k class visual signature of expression said inquiry image and said i width of cloth image;
Step 4, said Rough Inspection intelligence body carries out descending sort according to said comprehensive similarity, and with preceding M the preliminary search result { T that said comprehensive similarity is corresponding
j}
J=1 ..., MReturn to the intelligent body of said cooperation as input;
Step 5, the user carries out mark to the image in the said input according to preset quantitative evaluation grade, and said cooperation intelligence body feeds back to the intelligent body of smart inspection with user's mark result;
Step 6, the intelligent body of said smart inspection is adjusted image { T in said inquiry image and the said input according to user's mark result
j}
J=1 ..., MThe proper vector of all K class visual signatures between the shared weights of similarity, be designated as { W
Jk} '
J=1,2 ..., M; K=1 ..., K
Step 7, the intelligent body of said smart inspection recomputates the comprehensive similarity of all images in said inquiry image and the said remote sensing image storehouse according to following formula;
Step 8, the intelligent body of said smart inspection carries out descending sort again according to the comprehensive similarity after calculating, and result for retrieval { T after the renewal that the comprehensive similarity after the said calculating is corresponding
j} '
J=1 ..., MReturn to the intelligent body of said cooperation;
Step 9 contrasts the said image number that satisfies height correlation in the image number that satisfies height correlation in the result for retrieval of back and the said input of upgrading, and whether judges difference greater than certain predetermined threshold value, if, then with the back result for retrieval of said renewal as input, even { T
j}
J=(1 ..., M)={ T
j} '
J=(1 ..., M), { W
Jk}
J=(1 ..., M), k=(1 ..., K)={ W
Jk} '
J=(1 ..., M), k=(1 ..., K), and return said step 5, if not, the intelligent body of then said cooperation is exported said renewal back result for retrieval as final result for retrieval.
2. remote sensing image search method according to claim 1 is characterized in that:
Said visual signature comprises color characteristic, textural characteristics and shape facility;
Said feature intelligent body comprises color characteristic intelligence body, textural characteristics intelligence body and shape facility intelligence body;
Said Rough Inspection intelligence body comprises color Rough Inspection intelligence body, texture Rough Inspection intelligence body and shape Rough Inspection intelligence body.
3. remote sensing image search method according to claim 1 is characterized in that, said step 1 comprises following substep:
(1) for any width of cloth image I, the intelligent body of said color characteristic, said textural characteristics intelligence body and parallel color characteristic, textural characteristics and the shape facility that extracts said image of said shape facility intelligence body, specifically implementation does,
The hue histogram that said color characteristic intelligence body extracts said image and saturation degree histogram are to describe the color characteristic of said image, and the proper vector expression of said color characteristic as follows;
FC
I=(h
1,s
1,h
2,s
2,...,h
L,s
L)
Wherein, FC
IThe proper vector of representing said color characteristic, L representes the quantized interval size of said image tone component and saturation degree component in tone-saturation degree color space, h
lAnd s
lRepresent l said tone component and l the probability that said saturation degree component occurs in whole quantification space respectively, and l=(1,2 ..., L);
Said textural characteristics intelligence body at first carries out wavelet decomposition to said image, adopts the average of each wavelet sub-band and the textural characteristics that standard deviation is described said image then, and the proper vector of said textural characteristics is expressed as follows;
Wherein, FT
IThe proper vector of representing said textural characteristics, L representes high fdrequency component sum that said image is carried out obtaining after the wavelet decomposition,
With
Wavelet coefficient normalization average value and the standard deviation of representing l said high fdrequency component respectively, and l=(1,2 ..., L);
Said shape facility intelligence body at first adopts edge detection operator to extract edge images to said image, utilizes invariant moments to describe the shape facility of said image then, and the proper vector of said shape facility is expressed as follows:
Wherein, FS
IThe proper vector of representing said shape facility,
L Hu invariant moments group component of the edge images that expression employing Canny edge detection operator obtains, l=(1,2 ...., 7).
(2) proper vector of the said color characteristic of comprehensive said image, said textural characteristics and said shape facility is formed the comprehensive visual signature of said image according to following formula.
F
I={FC
I,FT
I,FS
I}
4. remote sensing image search method according to claim 1 is characterized in that, said step 3 comprises following substep:
(1) for any two width of cloth image I and J; Color characteristic similarity, textural characteristics similarity and the shape facility similarity of said color Rough Inspection intelligence body, said texture Rough Inspection intelligence body and said two width of cloth images of said shape Rough Inspection intelligence body parallel computation; Concrete implementation does
Said color Rough Inspection intelligence body calculates the color characteristic similarity of said two width of cloth images according to following formula,
Wherein,
and
representes l frequency that the tone component occurs at whole quantized interval in said two width of cloth images respectively;
and
representes the frequency that l saturation degree component of said two width of cloth images occurs at whole quantized interval respectively, and L is the quantized interval size; For the value that guarantees said color characteristic similarity between 0 to 1, need carry out Gaussian normalization according to following formula to said color characteristic similarity,
Wherein, μ and σ represent color characteristic similarity average and the standard deviation that normalization is preceding respectively;
Said texture Rough Inspection intelligence body calculates the textural characteristics similarity of said two width of cloth images according to following formula,
Wherein,
and
representes the normalization average value and the standard deviation of l high fdrequency component after said two width of cloth image wavelet decomposition respectively, and L is the high fdrequency component sum that wavelet decomposition obtains; For the value that guarantees said textural characteristics similarity between 0 to 1, need carry out Gaussian normalization according to following formula to said textural characteristics similarity,
Said shape Rough Inspection intelligence body calculates the shape facility similarity of said two width of cloth images according to following formula,
Wherein, l corresponding Hu invariant moments group component of said two width of cloth images of
and
expression; For the value that guarantees said shape facility similarity between 0 to 1, need carry out Gaussian normalization according to following formula to said shapes textures characteristic similarity,
(2) the intelligent body of said smart inspection is given initial weight W respectively to said color characteristic similarity, said textural characteristics similarity and said shape facility similarity
C, W
TAnd W
S, and calculate the comprehensive similarity of said two width of cloth images according to following formula.
Sim(I,J)=W
CSim
C′(I,J)+W
TSim
T′(I,J)+W
SSim
S′(I,J)
Wherein, W
C, W
TAnd W
SSatisfy W
C+ W
T+ W
S=1.
5. remote sensing image search method according to claim 1 is characterized in that: quantitative evaluation grade preset in the said step 5 comprises height correlation, is correlated with, does not select, has nothing to do and highly irrelevant five grades, and corresponding value is as follows,
Wherein, i=(1,2 ..., 5).
6. remote sensing image search method according to claim 1 is characterized in that, said step 6 comprises following substep:
Wherein, W
Jk' and W
JkRepresent respectively the intelligent body of said smart inspection according to user's mark result to said color characteristic, said textural characteristics and said shape facility in said comprehensive similarity shared weight after adjustment and the value before the adjustment, { T
j}
J=1 ..., MRepresent the similar image set that the intelligent body of said cooperation returns, { T
JC}
J=1,2 ..., M, { T
JT}
J=1,2 ..., M{ T
JS}
J=1,2 ..., MRepresent respectively to gather according to the similar image that said color characteristic similarity, said textural characteristics similarity and said shape facility similarity obtain separately.As the W that calculates according to following formula
Jk, make W at '<0 o'clock
Jk'=0;
(2) according to following formula weights are carried out normalization:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2011104247758A CN102591918A (en) | 2011-12-16 | 2011-12-16 | Remote sensing image retrieval method based on multi-agent system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2011104247758A CN102591918A (en) | 2011-12-16 | 2011-12-16 | Remote sensing image retrieval method based on multi-agent system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN102591918A true CN102591918A (en) | 2012-07-18 |
Family
ID=46480568
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2011104247758A Pending CN102591918A (en) | 2011-12-16 | 2011-12-16 | Remote sensing image retrieval method based on multi-agent system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN102591918A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108629337A (en) * | 2018-06-11 | 2018-10-09 | 深圳市益鑫智能科技有限公司 | A kind of face recognition door control system based on block chain |
CN109522904A (en) * | 2018-09-30 | 2019-03-26 | 广州地理研究所 | A kind of regular farmland extracting method based on remotely-sensed data |
CN110309804A (en) * | 2019-07-08 | 2019-10-08 | 上海天诚比集科技有限公司 | The reduction light for assigning power algorithm based on nine grids interferes image similarity detection method |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7194453B2 (en) * | 2000-03-29 | 2007-03-20 | Canon Kabushiki Kaisha | Image retrieval method and apparatus |
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
- 2011-12-16 CN CN2011104247758A patent/CN102591918A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7194453B2 (en) * | 2000-03-29 | 2007-03-20 | Canon Kabushiki Kaisha | Image retrieval method and apparatus |
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 |
Non-Patent Citations (1)
Title |
---|
程起敏: "基于内容的遥感影像库检索关键技术研究", 《中国博士学位论文全文数据库,信息科技辑》 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108629337A (en) * | 2018-06-11 | 2018-10-09 | 深圳市益鑫智能科技有限公司 | A kind of face recognition door control system based on block chain |
CN109522904A (en) * | 2018-09-30 | 2019-03-26 | 广州地理研究所 | A kind of regular farmland extracting method based on remotely-sensed data |
CN110309804A (en) * | 2019-07-08 | 2019-10-08 | 上海天诚比集科技有限公司 | The reduction light for assigning power algorithm based on nine grids interferes image similarity detection method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103413151B (en) | Hyperspectral image classification method based on figure canonical low-rank representation Dimensionality Reduction | |
CN102254303B (en) | Methods for segmenting and searching remote sensing image | |
CN101551823B (en) | Comprehensive multi-feature image retrieval method | |
CN103871029B (en) | A kind of image enhaucament and dividing method | |
CN101937079A (en) | Remote sensing image variation detection method based on region similarity | |
CN101556600B (en) | Method for retrieving images in DCT domain | |
CN102122352B (en) | Characteristic value distribution statistical property-based polarized SAR image classification method | |
CN102722883B (en) | Polarized SAR (synthetic aperture radar) image segmentation method with space adaptivity | |
CN104103082A (en) | Image saliency detection method based on region description and priori knowledge | |
CN101763440B (en) | Method for filtering searched images | |
CN102013017A (en) | Method for roughly sorting high-resolution remote sensing image scene | |
Guo et al. | City-wide canopy cover decline due to residential property redevelopment in Christchurch, New Zealand | |
CN106126585A (en) | Unmanned plane image search method based on quality grading with the combination of perception Hash feature | |
CN102542293A (en) | Class-I extraction and classification method aiming at high-resolution SAR (Synthetic Aperture Radar) image scene interpretation | |
CN102314610B (en) | Object-oriented image clustering method based on probabilistic latent semantic analysis (PLSA) model | |
Chen et al. | Clustering based on eigenspace transformation–CBEST for efficient classification | |
CN102163285A (en) | Cross-domain video semantic concept detection method based on active learning | |
CN102938069A (en) | Pure and mixed pixel automatic classification method based on information entropy | |
CN102591918A (en) | Remote sensing image retrieval method based on multi-agent system | |
Li et al. | Integrating multiple textural features for remote sensing image change detection | |
CN109740504B (en) | Method for extracting sea area resources based on remote sensing image | |
CN103345636B (en) | Based on the copper flotation site bubble condition recognition methods of multi-scale wavelet binaryzation | |
CN104361339A (en) | Palm image extracting and identification method | |
CN102609721B (en) | Remote sensing image clustering method | |
CN108985346A (en) | Fusion low layer pictures feature and showing for CNN feature survey image search method |
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: 20120718 |