CN110119716A - A kind of multi-source image processing method - Google Patents
A kind of multi-source image processing method Download PDFInfo
- Publication number
- CN110119716A CN110119716A CN201910400558.1A CN201910400558A CN110119716A CN 110119716 A CN110119716 A CN 110119716A CN 201910400558 A CN201910400558 A CN 201910400558A CN 110119716 A CN110119716 A CN 110119716A
- Authority
- CN
- China
- Prior art keywords
- source
- feature
- cluster
- image
- invariance
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000003672 processing method Methods 0.000 title claims abstract description 27
- 238000012805 post-processing Methods 0.000 claims abstract description 8
- 239000011159 matrix material Substances 0.000 claims description 52
- 239000000284 extract Substances 0.000 claims description 18
- 238000010276 construction Methods 0.000 claims description 14
- 230000000295 complement effect Effects 0.000 claims description 11
- 238000012804 iterative process Methods 0.000 claims description 3
- 230000004927 fusion Effects 0.000 abstract description 10
- 238000000605 extraction Methods 0.000 abstract description 6
- 238000000034 method Methods 0.000 description 23
- 230000008569 process Effects 0.000 description 8
- 230000008901 benefit Effects 0.000 description 5
- 238000012549 training Methods 0.000 description 4
- 238000000354 decomposition reaction Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 241000282326 Felis catus Species 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 241000894007 species Species 0.000 description 2
- XCWPUUGSGHNIDZ-UHFFFAOYSA-N Oxypertine Chemical compound C1=2C=C(OC)C(OC)=CC=2NC(C)=C1CCN(CC1)CCN1C1=CC=CC=C1 XCWPUUGSGHNIDZ-UHFFFAOYSA-N 0.000 description 1
- 238000000429 assembly Methods 0.000 description 1
- 230000000712 assembly Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000009792 diffusion process Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000011521 glass Substances 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 230000026676 system process Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
- G06V10/464—Salient features, e.g. scale invariant feature transforms [SIFT] using a plurality of salient features, e.g. bag-of-words [BoW] representations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the present invention provides a kind of multi-source image processing method, comprising: the first cluster category of several associated pictures is obtained using automatic cluster;Wherein, several described associated pictures include several images relevant to place or target obtained using multiple sensors;The Invariance feature and otherness feature of several associated pictures are at least extracted based on the first cluster category;And post processing of image is carried out according to the Invariance feature and the otherness feature, wherein described image post-processing includes target identification or image co-registration.The present invention comprehensively utilizes multi-source Remote Sensing Images, from the Invariance feature and otherness feature on the extraction of data itself, the different levels for interpreting different sensor images, different scale in the case where no priori.The present invention can be widely applied in multi-source Remote Sensing Image Fusion and target identification.
Description
Technical field
The technical fields such as the present invention relates to multi-source Remote Sensing Image Fusion, features to understand, expert's interpretation, and in particular to Yi Zhongduo
Source images processing method.
Background technique
The key of multi-source Remote Sensing Images feature interpretation is to extract the shared Invariance feature and difference of multi-source Remote Sensing Images
Complementary feature.Due to the difference of imaging mechanism, the shared Invariance feature and difference-complementary of multi-source Remote Sensing Images are automatically extracted
Property feature is extremely difficult, seriously constrains the practical application of multi-source heterogeneous remote sensing images.
Summary of the invention
In order to solve the above-mentioned technical problem, the invention proposes a kind of multi-source image processing methods.
The present invention proposes a kind of multi-source image processing method, comprising: obtains the of several associated pictures using automatic cluster
One cluster category;Wherein, several described associated pictures include several obtained using multiple sensors and place or target phase
The image of pass;Invariance feature and the otherness spy of several associated pictures are at least extracted based on the first cluster category
Sign;And post processing of image is carried out according to the Invariance feature and the otherness feature, wherein described image post-processing
Including target identification or image co-registration.
In some embodiments, the multi-source image processing method further include: generated based on several described associated pictures
Synthetic image obtains the second cluster category and multi-source invariance sample set of the synthetic image using automatic cluster,
In, the second cluster category includes the classification of cluster centre and each pixel;Based on several described associated pictures using automatic
Cluster obtains multi-source otherness sample set, according to the second cluster category resetting the first cluster category;And it is based on
The multi-source otherness sample set, the multi-source invariance sample set, the first cluster category, the second cluster class
It is marked with and feature explains equation, extract the Invariance feature and otherness feature of the multiple associated picture.
In some embodiments, the multi-source image processing method further includes, based on several described associated pictures and
The invariable features extraction residual image, and multiple dimensioned multi-level features interpretation carried out to the residual image, described in extraction
The Invariance feature of residual image and complementary feature.
In some embodiments, the multiple dimensioned multi-level features interpretation is above mentioned in the remaining multi-source image of current level
Take the iterative process of Invariance feature and complementary feature.
In some embodiments, several described associated pictures are the multi-source Remote Sensing Images obtained using multiple sensors.
In some embodiments, the similitude and otherness of the common trait based on the multi-source Remote Sensing Images are to described comprehensive
It closes image and carries out automatic cluster.
In some embodiments, the multi-source Remote Sensing Images include several single source images;Phase based on single source images
Automatic cluster is carried out to the multi-source Remote Sensing Images like property and otherness.
In some embodiments, single source images are gathered on the basis of the cluster centre of the synthetic image and category
Class category is reset.
In some embodiments, it is mutual to be based on multi-source Invariance feature Laplacian Matrix, multi-source for the feature interpretation equation
Benefit property feature Laplacian Matrix and isotonicity feature Laplacian Matrix construction.
In some embodiments, the multi-source image processing method, further includes: be based on the multi-source invariance sample set
It closes and its category, the multi-source otherness sample set and its category construction multi-source Invariance feature matrix, multi-source is complementary special
Levy matrix and isotonicity eigenmatrix;It solves the feature interpretation equation and obtains generalized eigenvalue and generalized eigenvector;By
The corresponding generalized eigenvector of non-zero minimum generalized eigenvalue of the feature interpretation equation generates feature and interprets projection matrix;With
And the Invariance feature and otherness feature of several associated pictures are extracted based on feature interpretation projection matrix.
The method of the invention is for the feature invariance and otherness of interpretation multi-source Remote Sensing Images, understanding and solves multi-source
The difficult point of data fusion and target identification has great importance, and major advantage is as follows:
The present invention considers the complexity and versatility of multi-source Remote Sensing Images feature interpretation, utilizes multi-source image self information
Feature invariance and otherness are excavated, training sample and priori knowledge are not needed, there is no limit have wide to image type number
General universality.
The present invention fully considered feature invariance and otherness over the ground species not, the dependence of scale, it is distant by multi-source
Similitude and otherness construction feature interpretation equation of the sense image based on cluster category and it is progressive it is iterative extract different levels,
The feature invariance and otherness of different scale, substantially increase feature interpretation capability and performance.
Owing to the above advantages, present invention greatly enhances multi-source Remote Sensing Images to interpret performance, to multi-source Remote Sensing Images
Fusion and feature interpretation provide good support, can be widely applied to multi-source Remote Sensing Image Fusion, target identification, scene classification
Etc. in systems.
Detailed description of the invention
Fig. 1 is the multi-source image processing method that an embodiment of the present invention provides.
Fig. 2 is the multi-source image processing method that an embodiment of the present invention provides.
The multi-source Remote Sensing Images feature that Fig. 3 further embodiment of this invention provides interprets flow chart.
Specific embodiment
Below in conjunction with attached drawing, the technical solution in the embodiment of the present disclosure is clearly and completely described with reference to attached
The non-limiting example embodiment for showing in figure and being described in detail in the following description, the example that the disclosure is more fully described below are implemented
Example and their various features and Advantageous details.It should be noted that feature shown in figure is not required to be drawn to scale.This
The open description that known tip assemblies and technology is omitted, to not make the example embodiment of the disclosure fuzzy.Given example is only
It is intended to be conducive to understand the implementation of disclosure example embodiment, and those skilled in the art is further enable to implement example reality
Apply example.Thus, these examples are understood not to the limitation to the range of embodiment of the disclosure.
Unless otherwise specifically defined, the technical term or scientific term that the disclosure uses should be disclosure fields
The ordinary meaning that the interior personage with general technical ability is understood." first ", " second " used in the disclosure and similar word
Language is not offered as any sequence, quantity or importance, and is used only to distinguish the image of different features.In addition, in this public affairs
It opens in each embodiment, same or similar reference label indicates same or similar component.
There is very strong complementarity between multi-source heterogeneous remote sensing images, comprehensively utilize the complementarity of multi-source heterogeneous remote sensing images
Play a significant role to the performance for improving target identification, multi-source Remote Sensing Images feature is interpreted in multi-source Remote Sensing Image Fusion, target
Numerous application fields such as identification, feature understanding have urgent application demand.
The embodiment of the present invention is from multi-source Remote Sensing Images, in the case where not depending on training sample or expertise
The Invariance feature and complementarity feature for automatically extracting multi-source Remote Sensing Images, can be multi-source Remote Sensing Image Fusion, target identification
Etc. practical applications reliable decision-making foundation is provided.
For the difficult point of multi-source Remote Sensing Images feature interpretation and the demand of practical application, provide one kind has the embodiment of the present invention
The multi-source Remote Sensing Images feature decomposition method of effect.
As shown in Figure 1, the present invention provides a kind of multi-source image processing method.The image processing method includes: step 101,
The first cluster category of several associated pictures is obtained using automatic cluster;Wherein, several described associated pictures include using multiple
Several images relevant to place or target that sensor obtains;Step 102, it is at least extracted based on the first cluster category
The Invariance feature and otherness feature of several associated pictures;And step 103, according to the Invariance feature and institute
It states otherness feature and carries out post processing of image, wherein described image post-processing includes target identification or image co-registration.For example, step
Rapid 101 several associated pictures being related to can be for using the multi-source Remote Sensing Images of multiple sensors acquisition.In some instances, base
Automatic cluster is carried out to the synthetic image in the similitude and otherness of the common trait of the multi-source Remote Sensing Images.
Several associated pictures are handled by the way of automatic cluster can be from several associated pictures (for example, multi-source remote sensing figure
Picture) it sets out in itself, several associated picture (multi-source remote sensing figures are automatically extracted in the case where not depending on training sample or expertise
Picture) Invariance feature and complementary feature, can be provided reliably for practical applications such as multi-source Remote Sensing Image Fusion, target identifications
Decision-making foundation.
As shown in Fig. 2, in some embodiments, multi-source image processing method further include: step 201, based on it is described several
Associated picture generates synthetic image, and the second cluster category and multi-source invariance of the synthetic image are obtained using automatic cluster
Sample set, wherein the second cluster category includes the classification of cluster centre and each pixel;Step 101 further includes being based on
Several described associated pictures obtain multi-source otherness sample set using automatic cluster, reset institute according to the second cluster category
State the first cluster category;And it includes being based on the multi-source otherness sample set, the multi-source invariance that step 102 is exemplary
Sample set, the first cluster category, the second cluster category and feature explain equation, extract several described correlation figures
The Invariance feature and otherness feature of picture.
In some embodiments, multi-source image processing method further includes, based on several described associated pictures and it is described not
Vertic features extract residual image, and carry out multiple dimensioned multi-level features interpretation to the residual image, extract the remaining figure
The Invariance feature of picture and complementary feature.For example, the multiple dimensioned multi-level features interpretation may include in current level
The iterative process of Invariance feature and complementary feature is extracted on remaining multi-source image.
In some embodiments, the multi-source Remote Sensing Images include several single source images;Phase based on single source images
Automatic cluster is carried out to the multi-source Remote Sensing Images like property and otherness.For example, in some instances with the cluster of synthetic image
The cluster category of single source images is reset on the basis of center and category.
In some embodiments, it is mutual to be based on multi-source Invariance feature Laplacian Matrix, multi-source for the feature interpretation equation
Benefit property feature Laplacian Matrix and isotonicity feature Laplacian Matrix construction.For example, multi-source image processing method includes:
It is constant based on the multi-source invariance sample set and its category, the multi-source otherness sample set and its category construction multi-source
Property eigenmatrix, multi-source complementarity eigenmatrix and isotonicity eigenmatrix;It solves the feature interpretation equation and obtains broad sense
Characteristic value and generalized eigenvector;By the corresponding generalized eigenvector of non-zero minimum generalized eigenvalue of feature interpretation equation
It generates feature and interprets projection matrix;And the invariance of several associated pictures is extracted based on feature interpretation projection matrix
Feature and otherness feature.
The present invention also provides the embodiments of multi-source Remote Sensing Images processing, as shown in Figure 3.Multi-source remote sensing figure of the invention
As processing method includes the following steps:
Step S1 generates synthetic image by multi-source Remote Sensing Images, obtains cluster class by automatic cluster on synthetic image
Mark, multi-source invariance sample.
Step S2 obtains cluster category, multi-source otherness sample by automatic cluster respectively on single source images, and with comprehensive
The cluster category of single source images is reset on the basis of the cluster centre and category of conjunction image.
Step S3 is based on multi-source invariance sample, multi-source otherness sample and its category and constructs multi-source invariance characteristic matrix
Battle array, multi-source complementarity eigenmatrix and isotonicity eigenmatrix solve feature interpretation equation and obtain generalized eigenvalue and broad sense
Feature vector and construction feature interpretation projection matrix extraction Invariance feature and otherness feature.
Step S4 is based on multi-source Remote Sensing Images and invariant features image configuration remnants multi-source image, and in remaining multi-source image
Upper iterative step S2 and step S3 extracts the Invariance feature and otherness feature of different levels, different scale, changes until meeting
For termination condition.
With reference to Fig. 3, for above-mentioned steps, the present invention also provides following examples to be illustrated.
Step S1 generates synthetic image by multi-source Remote Sensing Images, obtains cluster class by automatic cluster on synthetic image
Mark, multi-source invariance sample.Detailed process is as follows for the synthetic image and automatic cluster.
Step S11 synthetic image generates.Enable IkIndicate the multi-source Remote Sensing Images of the Same Scene shot by different sensors,
And multi-source Remote Sensing Images are registered, wherein K is the species number of multi-source Remote Sensing Images.The imaging mechanism of multi-source Remote Sensing Images is not
Together, feature difference is very big, and wave band number is not also identical.For example, I1For full-colour image, I2For multispectral image, I3For infrared image,
I4For SAR image, I5For high spectrum image.In order to avoid the classification as caused by wave band quantity variance is unbalance, multi-band image exists
Mean value is taken to obtain gray level image in wave band dimension.Specifically, if multi-source Remote Sensing Images IkWave band number be 1, then multi-source remote sensing figure
As IkCorresponding average gray image is Gk=Ik;If multi-source Remote Sensing Images IkWave band number be greater than 1, then multi-source Remote Sensing Images IkIt is right
The average gray image answered is Gk=mean (3, Ik), wherein (3, I meank) indicate to multi-source Remote Sensing Images IkIn wave band dimension
On take average operation.Synthetic image X is linked in wave band dimension by single band image or average gray image, i.e. X=cat
(3,G1,G2..., GK), and cat (3, G1,G2..., GK) indicate to gray level image G1,G2..., GKChain is carried out in wave band dimension
Connect operation.
Step S12 automatic cluster.This example is a kind of unsupervised multi-source Remote Sensing Images feature decomposition method, without priori
Knowledge or training sample can use.For this purpose, this example by synthetic image X automatic cluster extract the phase of multi-source image
Like property and otherness.Automatic cluster is using the related coefficient of the multiband gray feature vector of each pixel of synthetic image X as similar
Property measurement, all pixels are regarded as potential cluster centre point in initialization, then pass through iterative diffusion sense of duty message
The cluster centre of each classification is found with availability message and determines the packet class of each pixel.The sense of duty message r (i, k) table
Show that k-th of pixel is suitable as the degree of the cluster centre of ith pixel.Availability message a (i, k) indicates ith pixel choosing
Select a possibility that k-th of pixel is as its cluster centre.Detailed process is as follows for automatic cluster:
S121 original state: availability message a (i, k)=0;
Sense of duty message S122 all according to availability information updating, i.e.,
Wherein, s (i, k) indicates mutual between the multiband gray feature vector between ith pixel and k-th of pixel
Relationship number.
S123 updates all availability message, i.e., according to sense of duty message
S124 combination availability message and sense of duty message determine cluster centre.For ith pixel, if can be used
Property message and sense of duty message and the i.e. k of " a (i, k)+r (i, k) " when being maximized equal to i, then pixels illustrated i itself is poly-
Class center;If k is unequal with i, pixels illustrated i is attachment point, and cluster centre is pixel k.
If S125 reaches the message variable quantity in the maximum number of iterations T or data point of setting and is less than given threshold tau,
Algorithm terminates;Otherwise, S122 step is gone to.In this example, the number of iterations T=100, threshold tau=10.Those skilled in the art can
To set the specific trend of the number of iterations T and the specific value of threshold tau according to actual needs.
At the end of S13 automatic cluster algorithm iteration, the cluster centre c of available synthetic image Xl(l=1 ..., L) and
The classification of each pixel.For convenience of narration, above-mentioned cluster process is denoted as f0.For any pixel p, f of synthetic image X0(p) table
Show the corresponding cluster category of pixel p.Synthetic image X possesses the collection of the corresponding multiband gray feature vector of pixel of the same category
It is collectively referred to as such multi-source invariance sample set.There is L cluster classification just to have L multi-source invariance sample set.
Step S2 obtains cluster category, multi-source otherness sample by automatic cluster respectively on single source images, and with comprehensive
The cluster category of single source images is reset on the basis of the cluster centre and category of conjunction image.The list source images automatic cluster
And detailed process is as follows for category resetting:
The mono- source images automatic cluster of step S21.Respectively with single source images IkIt substitutes synthetic image X and utilizes step S12 institute
The method stated is to single source images IkAutomatic cluster is carried out, single source cluster device f of each type image is obtainedk, cluster centreWith it is every
The classification of a pixel.Equally, according to the available single source images I of method described in step S13kCategory sample.It chats for convenience
It states, usesIndicate synthetic image X, single source images IkOn i-th of sample, category useIt indicates, wherein k=0 ...,
K.
The resetting of step S22 category.Since the cluster process of synthetic image and multi-source image independently carries out, their class
Mark lacks correspondence.For this purpose, to single source images I on the basis of the cluster centre of synthetic image X and categorykCluster category carry out
Resetting.Specific practice are as follows: by image IkPixel p category resetting are as follows:
Wherein, s (ck(p),c0)) indicate image IkPixel p cluster centre ck(p) in image IkOn whether there is also it
Its cluster centre corresponds to the same cluster centre c on synthetic image X simultaneously0(ck(p))。
Step S3 is based on multi-source invariance sample, multi-source otherness sample and its category and constructs multi-source invariance characteristic matrix
Battle array, multi-source complementarity eigenmatrix and isotonicity eigenmatrix solve feature interpretation equation and obtain generalized eigenvalue and broad sense
Feature vector and construction feature interpretation projection matrix extraction Invariance feature and otherness feature.Eigenmatrix mainly includes multi-source
Invariance feature matrix, multi-source complementarity eigenmatrix and isotonicity eigenmatrix.Every kind of eigenmatrix all includes basic square
Battle array, capable and diagonal matrix and Laplacian Matrix.The eigenmatrix, detailed process is as follows for the building of feature interpretation equation:
Step S31 multi-source Invariance feature matrix construction.Multi-source invariance basis matrix WsIt is made of, has block matrix
Body form are as follows:
It is a maRow mbThe matrix of column, maAnd mbRespectively indicate IaAnd IbIn number of samples.Indicate pixel
I corresponding cluster centre on synthetic image.ConditionIt is meant that image Ia
On pixel i and image IbOn pixel j classification it is different but they are identical in the corresponding classification of synthetic image X.Multi-source is constant
Property row and diagonal matrixMulti-source invariance Laplacian Matrix Ls=Ds-Ws。ForIt is corresponding
The matrix of a=0, b=0;ForThe matrix of corresponding a=0, b=K;ForThe square of corresponding a=K, b=0
Battle array;ForThe matrix of corresponding a=K, b=K;For image IaOn pixel i classification;For image Ib
On pixel j classification;For image IaOn pixel i in the corresponding classification of synthetic image X;For
Image IbOn pixel j in the corresponding classification of synthetic image X.
Step S32 multi-source complementarity eigenmatrix construction.Multi-source complementarity eigenmatrix WdAnd be made of block matrix,
Concrete form are as follows:
Multi-source complementarity row and diagonal eigenmatrixMulti-source complementarity Laplacian Matrix Ld=
Dd-Wd。ForThe matrix of corresponding a=0, b=0;ForThe matrix of corresponding a=0, b=K;For
The matrix of corresponding a=K, b=K;ForThe matrix of corresponding a=K, b=K.
Step S33 isotonicity eigenmatrix construction.List source isotonicity similarity matrix W is constructed firstk, single source isotonicity row
With diagonal eigenmatrix Dk, list source isotonicity Laplce Lk, the list source isotonicity similarity matrix Wk, single source isotonicity row
With diagonal eigenmatrix DkWith single source isotonicity Laplce LkEach element be calculated as follows: Lk=Dk-Wk。
Total topology relational matrix is
General characteristic matrix are
Wherein, I0=X, X are synthetic image described in above-mentioned steps S11.
Step S34 feature interprets equation building and the interpretation of multi-source feature.Feature interpretation is drawn based on the multi-source invariance
This matrix L of pulas, multi-source complementarity Laplacian Matrix Ld, total topology relational matrix L, general characteristic matrix Z carry out feature
It extracts and feature interpretation, feature interprets equation are as follows: Z (L+Ls)ZTμ=λ ZLdZTμ.It solves feature interpretation equation and obtains generalized character
Value λkWith generalized eigenvector μk.The smallest (K+1) a non-zero generalized eigenvalue λkCorresponding generalized eigenvector μkIt is characterized
Projecting direction.Specifically, to IkFeature vector x at middle ith pixeli, it is in IjIn be expressed asTable
Show μjPseudoinverse.With IjComplementary feature beI is unit matrix.xiGeneral character in different type image
Feature isμ0Indicate (K+1) a maximum λ of minimum non-zero generalized eigenvalue intermediate valuekCorresponding generalized eigenvector.
Step S4 is based on multi-source Remote Sensing Images and invariant features image configuration remnants multi-source image, and in remaining multi-source image
Upper iterative step S2 and step S3 extracts the Invariance feature and otherness feature of different levels, different scale, changes until meeting
For termination condition.The remnants multi-source image extracts, detailed process is as follows for fine-feature interpretation:
Step S41 remnants' multi-source image extracts.In order to understand on finer scale multi-source Remote Sensing Images general character and
Complementarity first calculates the residual characteristics of a upper scale.Specifically, to IkThe corresponding feature vector x of each pixeli, calculate remaining
Feature vectorμ0Indicate (K+1) a maximum λ of minimum non-zero generalized eigenvalue intermediate valuekCorresponding broad sense
Feature vector, then by the residual characteristics vector x of different pixelsi' according to xiOriginal column locations form new image I'k。
The interpretation of step S42 fine-feature.Then according to step S2 to step S3 the method in new multi-source image I'kAbove mention
It takes multi-source otherness sample, construction feature matrix, extract Invariance feature and complementary feature.
The switching of step S43 scale.If I'kEach pixel characteristic vector quadratic sum IkEach pixel characteristic vector it is flat
The ratio of side's sum is both less than ε, then iteration ends;Otherwise, scale switching and feature interpretation are carried out always until meeting iteration ends
The specific practice of condition, scale switching and feature interpretation is that remaining multi-source image I' is updated according to step S41 the methodk, press
Fine-feature interpretation is carried out according to step S42 the method.
The benefit of multi-level features interpretation is gradually to understand multi-source Remote Sensing Images in different layers by way of " magnifying glass "
Invariance and otherness secondary, on different scale.In this way, the general character invariant features and complementary dif-ference feature of multi-source Remote Sensing Images
It just completely, in all directions extracts, this is very important the subsequent applications such as Fusion Features, target identification.
To sum up, it includes: step S1 that the embodiment of the present invention, which provides a kind of multi-source Remote Sensing Images feature decomposition method: distant by multi-source
Feel image and generate synthetic image, cluster category, multi-source invariance sample are obtained by automatic cluster on synthetic image.Step
S2: cluster category, multi-source otherness sample are obtained by automatic cluster respectively on single source images, and with the cluster of synthetic image
The cluster category of single source images is reset on the basis of center and category.Step S3: poor based on multi-source invariance sample, multi-source
Anisotropic sample and its category construction multi-source Invariance feature matrix, multi-source complementarity eigenmatrix and isotonicity eigenmatrix,
It solves feature interpretation equation and obtains generalized eigenvalue and generalized eigenvector and construction feature interpretation projection matrix extraction invariance
Feature and otherness feature.Step S4: it is based on multi-source Remote Sensing Images and invariant features image configuration remnants multi-source image, and residual
Iterative step S2 and step S3 extracts the Invariance feature and otherness feature of different levels, different scale on remaining multi-source image,
Until meeting stopping criterion for iteration.The present invention comprehensively utilizes multi-source Remote Sensing Images, goes out in the case where no priori from data itself
Hair extracts, interprets different levels, the Invariance feature on different scale and the otherness feature of different sensor images.The present invention
It can be widely applied in multi-source Remote Sensing Image Fusion and target identification.
Those skilled in the art should be able to recognize that, mould described in conjunction with the examples disclosed in the embodiments of the present disclosure
Block, method and step, can be realized with electronic hardware, computer software, or a combination of the two, software module, method and step pair
The program answered can be placed in random access memory (RAM), memory, read-only memory (ROM), electrically programmable ROM, electric erasable and can compile
Any other form of storage well known in journey ROM, register, hard disk, moveable magnetic disc, CD-ROM or technical field is situated between
In matter.In order to clearly demonstrate the interchangeability of electronic hardware and software, in the above description according to function generally
Describe each exemplary composition and step.These functions are executed actually with electronic hardware or software mode, depend on technology
The specific application and design constraint of scheme.Those skilled in the art can carry out using distinct methods each specific application
Realize described function, but such implementation should not be considered as beyond the scope of the present invention.
Term " includes " or any other like term are intended to cover non-exclusive inclusion, so that including a system
Process, method, article or equipment/device of column element not only includes those elements, but also including being not explicitly listed
Other elements, or further include the intrinsic element of these process, method, article or equipment/devices.
So far, it has been combined preferred embodiment shown in the drawings and describes technical solution of the present invention, still, this field
Technical staff is it is easily understood that protection scope of the present invention is expressly not limited to these specific embodiments.Without departing from this
Under the premise of the principle of invention, those skilled in the art can make equivalent change or replacement to the relevant technologies feature, these
Technical solution after change or replacement will fall within the scope of protection of the present invention.
Claims (10)
1. a kind of multi-source image processing method, comprising:
The first cluster category of several associated pictures is obtained using automatic cluster;Wherein, several described associated pictures include using
Several images relevant to place or target that multiple sensors obtain;
The Invariance feature and otherness feature of several associated pictures are at least extracted based on the first cluster category;And
Post processing of image is carried out according to the Invariance feature and the otherness feature, wherein described image post-processing packet
Include target identification or image co-registration.
2. multi-source image processing method as described in claim 1, further includes:
Synthetic image is generated based on several described associated pictures, the second cluster class of the synthetic image is obtained using automatic cluster
It is marked with and multi-source invariance sample set, wherein the second cluster category includes the classification of cluster centre and each pixel;
Multi-source otherness sample set is obtained using automatic cluster based on several described associated pictures, according to the second cluster class
Indicated weight sets the first cluster category;And
Based on the multi-source otherness sample set, the multi-source invariance sample set, the first cluster category, described the
Two cluster categories and feature explain equation, extract the Invariance feature and otherness feature of several associated pictures.
3. multi-source image processing method as claimed in claim 2, further includes, based on several described associated pictures and it is described not
Vertic features extract residual image, and carry out multiple dimensioned multi-level features interpretation to the residual image, extract the remaining figure
The Invariance feature of picture and complementary feature.
4. multi-source image processing method according to claim 3, wherein the multiple dimensioned multi-level features interpretation is to work as
The iterative process of Invariance feature and complementary feature is extracted on the remaining multi-source image of preceding level.
5. multi-source image processing method as claimed in claim 1,2 or 3, wherein several described associated pictures are using multiple
The multi-source Remote Sensing Images that sensor obtains.
6. multi-source image processing method as claimed in claim 5, wherein the common trait based on the multi-source Remote Sensing Images
Similitude and otherness carry out automatic cluster to the synthetic image.
7. multi-source image processing method as claimed in claim 6, wherein the multi-source Remote Sensing Images include several Dan Yuantu
Picture;
Similitude and otherness based on single source images carry out automatic cluster to the multi-source Remote Sensing Images.
8. multi-source image processing method as claimed in claim 7, wherein be designated as with the cluster centre and class of the synthetic image
Benchmark resets the cluster category of single source images.
9. multi-source image processing method as claimed in claim 7, wherein it is special that the feature interpretation equation is based on multi-source invariance
Levy Laplacian Matrix, multi-source complementarity feature Laplacian Matrix and isotonicity feature Laplacian Matrix construction.
10. multi-source image processing method as claimed in claim 9, further includes,
Multi-source is constructed based on the multi-source invariance sample set and its category, the multi-source otherness sample set and its category
Invariance feature matrix, multi-source complementarity eigenmatrix and isotonicity eigenmatrix;
It solves the feature interpretation equation and obtains generalized eigenvalue and generalized eigenvector;
Feature interpretation projection is generated by the corresponding generalized eigenvector of non-zero minimum generalized eigenvalue of feature interpretation equation
Matrix;And
The Invariance feature and otherness feature of several associated pictures are extracted based on feature interpretation projection matrix.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910400558.1A CN110119716B (en) | 2019-05-15 | 2019-05-15 | Multi-source image processing method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910400558.1A CN110119716B (en) | 2019-05-15 | 2019-05-15 | Multi-source image processing method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110119716A true CN110119716A (en) | 2019-08-13 |
CN110119716B CN110119716B (en) | 2021-05-11 |
Family
ID=67522433
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910400558.1A Active CN110119716B (en) | 2019-05-15 | 2019-05-15 | Multi-source image processing method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110119716B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114549883A (en) * | 2022-02-24 | 2022-05-27 | 北京百度网讯科技有限公司 | Image processing method, deep learning model training method, device and equipment |
GB2622776A (en) * | 2022-09-23 | 2024-04-03 | Continental Automotive Gmbh | Method and system for associating two or more images |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102902979A (en) * | 2012-09-13 | 2013-01-30 | 电子科技大学 | Method for automatic target recognition of synthetic aperture radar (SAR) |
CN104408751A (en) * | 2014-10-27 | 2015-03-11 | 中国科学院自动化研究所 | Hyperspectral image in-orbit compression method |
CN106897730A (en) * | 2016-12-30 | 2017-06-27 | 陕西师范大学 | SAR target model recognition methods based on fusion classification information with locality preserving projections |
-
2019
- 2019-05-15 CN CN201910400558.1A patent/CN110119716B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102902979A (en) * | 2012-09-13 | 2013-01-30 | 电子科技大学 | Method for automatic target recognition of synthetic aperture radar (SAR) |
CN104408751A (en) * | 2014-10-27 | 2015-03-11 | 中国科学院自动化研究所 | Hyperspectral image in-orbit compression method |
CN106897730A (en) * | 2016-12-30 | 2017-06-27 | 陕西师范大学 | SAR target model recognition methods based on fusion classification information with locality preserving projections |
Non-Patent Citations (3)
Title |
---|
JIXIAN ZHANG: "Multi-source remote sensing data fusion: status and trends", 《INTERNATIONAL JOURNAL OF IMAGE AND DATA FUSION》 * |
MISDETECTIONTIANCHENG LI等: "Multi-source homogeneous data clustering for multi-target detection from cluttered background with misdetection", 《APPLIED SOFT COMPUTING》 * |
雷琳: "多源遥感图像舰船目标特征提取与融合技术研究", 《中国博士学位论文全文数据库(电子期刊)INFORMATION SCIENCE AND TECHNOLOGY辑》 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114549883A (en) * | 2022-02-24 | 2022-05-27 | 北京百度网讯科技有限公司 | Image processing method, deep learning model training method, device and equipment |
CN114549883B (en) * | 2022-02-24 | 2023-09-05 | 北京百度网讯科技有限公司 | Image processing method, training method, device and equipment for deep learning model |
GB2622776A (en) * | 2022-09-23 | 2024-04-03 | Continental Automotive Gmbh | Method and system for associating two or more images |
Also Published As
Publication number | Publication date |
---|---|
CN110119716B (en) | 2021-05-11 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Berman et al. | Non-local image dehazing | |
Brown et al. | Multi-image matching using multi-scale oriented patches | |
Cao et al. | Landmark recognition with sparse representation classification and extreme learning machine | |
CN101138007B (en) | Image processing system, learning device and method | |
CN106447704B (en) | Visible light based on marking area feature and edge degree-infrared image method for registering | |
Benediktsson et al. | Classification and feature extraction for remote sensing images from urban areas based on morphological transformations | |
CN102629374B (en) | Image super resolution (SR) reconstruction method based on subspace projection and neighborhood embedding | |
CN110119716A (en) | A kind of multi-source image processing method | |
CN107742133A (en) | A kind of sorting technique for Polarimetric SAR Image | |
Kekre et al. | Image fusion using Kekre's hybrid wavelet transform | |
Liu et al. | Keypoint matching by outlier pruning with consensus constraint | |
Javed et al. | Multi-denoising based impulse noise removal from images using robust statistical features and genetic programming | |
Vasilyeva et al. | Lossy compression of three-channel remote sensing images with controllable quality | |
CN112329818A (en) | Hyperspectral image unsupervised classification method based on graph convolution network embedded representation | |
Nilubol et al. | Translational-and rotational-invariant hidden Markov model for automatic target recognition | |
Marpu et al. | Hyperspectral change detection using IR-MAD and feature reduction | |
Zhao et al. | Semi-supervised recurrent complex-valued convolution neural network for polsar image classification | |
Li et al. | A robust incremental learning framework for accurate skin region segmentation in color images | |
Liu et al. | Scene-level land use classification based on multi-features soft-probability cascading | |
Abadi et al. | Evaluation of feature points descriptors' performance for visual finger printing localization of smartphones | |
Lin et al. | Automatic detection and counting of small yellow thrips on lotus leaf back based on YOLO combined with VDSR and DPSR network | |
Yu et al. | High-resolution polsar classification model based on the fully convolutional neural network and polarimetric parameters | |
Alegavi et al. | Improving classification error for mixed pixels in satellite images using soft thresholding technique | |
Zhu et al. | Automatic remote sensing image registration based on SIFT descriptor and image classification | |
Forero et al. | Structured outlier models for robust dictionary learning |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |