CN105787943B - SAR image registration method based on multi-scale image block feature and rarefaction representation - Google Patents
SAR image registration method based on multi-scale image block feature and rarefaction representation Download PDFInfo
- Publication number
- CN105787943B CN105787943B CN201610118533.9A CN201610118533A CN105787943B CN 105787943 B CN105787943 B CN 105787943B CN 201610118533 A CN201610118533 A CN 201610118533A CN 105787943 B CN105787943 B CN 105787943B
- Authority
- CN
- China
- Prior art keywords
- point
- reference picture
- image
- registration
- characteristic point
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 45
- 238000010276 construction Methods 0.000 claims abstract description 6
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 claims abstract description 4
- 230000002159 abnormal effect Effects 0.000 claims abstract description 4
- 238000000354 decomposition reaction Methods 0.000 claims description 17
- 238000004422 calculation algorithm Methods 0.000 claims description 6
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 claims description 4
- 241000208340 Araliaceae Species 0.000 claims description 3
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims description 3
- 235000003140 Panax quinquefolius Nutrition 0.000 claims description 3
- 230000006978 adaptation Effects 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 3
- 235000008434 ginseng Nutrition 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 230000001351 cycling effect Effects 0.000 claims description 2
- 230000000694 effects Effects 0.000 abstract description 2
- 230000002708 enhancing effect Effects 0.000 abstract 1
- 238000005516 engineering process Methods 0.000 description 5
- 238000004088 simulation Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 230000007812 deficiency Effects 0.000 description 2
- 235000013399 edible fruits Nutrition 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000005856 abnormality Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000010287 polarization Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/14—Transformations for image registration, e.g. adjusting or mapping for alignment of images
- G06T3/147—Transformations for image registration, e.g. adjusting or mapping for alignment of images using affine transformations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
- G06T2207/10044—Radar image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20048—Transform domain processing
- G06T2207/20064—Wavelet transform [DWT]
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The SAR image registration method based on multi-scale image block feature and rarefaction representation that the invention discloses a kind of, mainly solve the problems, such as existing method for registering be applied to SAR image registration effect it is bad, realize the step of be:1) two width SAR images are inputted, an optional width, which is used as, refers to image, using another width as image subject to registration;2) reference picture characteristic point is chosen;3) feature point description of multiscale image block feature construction reference picture and image subject to registration is utilized to accord with;4) matching double points between reference picture and image subject to registration are established;5) abnormal point in matching double points is removed;6) according to finally obtained matching double points, affine Transform Model is established, geometric deformation parameter is obtained using least square method, obtains registration result.Compared with prior art, the present invention enhancing the robustness to speckle noise, the accuracy and registration accuracy of matching double points are improved, can be used for image co-registration and variation detects.
Description
Technical field
The invention belongs to technical field of image processing, and in particular to the method for registering images in radar image processing can be used
It is detected in image co-registration and variation.
Background technology
Synthetic aperture radar SAR system is round-the-clock because of its round-the-clock, has the characteristics that penetrability and is widely used in military affairs
With civilian neighborhood.For SAR image registration as the key link in SAR image application, it is to being derived from different time, different visual angles
Same scenery two width or several SAR images process for being matched, being superimposed.
For image registration problem, the method proposed at present can substantially be divided into two classes:Based on gray scale and feature based
Method for registering.Method for registering based on gray scale directly utilizes the half-tone information of image, by establishing certain phase between image pixel
Like property calculated measure come the registration parameters such as corresponding translation, rotation when searching out Optimum Matching.Most common matching based on gray scale
Quasi- method is the method for registering based on mutual information.Although this method comparison is intuitive, easy to implement, computation complexity is high, holds
It is easily absorbed in locally optimal solution, and easily affected by noise.The method for registering of feature based is due to being not directly placed on image ash
Angle value, but the feature for acting on image itself, thus have stronger adaptability to grey scale change, and calculation amount is small, it can
Handle the registration problems between image.The method for registering of most common feature based is to be based on Scale invariant features transform SIFT feature
Method for registering.However since there are speckle noises in SAR image, the method for registering of feature based is in processing SAR image registration
When, it is more likely that speckle noise is detected as characteristic point, to which a large amount of error matching points can be brought, leads to the registration knot of mistake
Fruit.
Invention content
It is an object of the invention to propose a kind of SAR image registration side based on multi-scale image block feature and rarefaction representation
Method causes registration accuracy is not high to ask to solve to carry out SAR image in the prior art with occurring a large amount of error matching points on time
Topic.
Realizing the technical thought of the object of the invention is:Highly reliable characteristic point is selected using spatial coherence, using more
Scale image block characteristics form feature descriptor, and the minimum difference criterion calculated according to rarefaction representation technology obtains best matching
Point pair, effectively enhances the robustness to speckle noise, and implementation step includes as follows:
(1) two images are inputted, an optional width, which is used as, refers to image I1, using another width as image I subject to registration2;
(2) reference picture characteristic point is chosen:
(2a) is using SIFT algorithms extraction reference picture I1Characteristic point, and by I1All characteristic points be stored in the first collection
It closes in R;
(2b) arbitrarily chooses a characteristic point r from reference picture set of characteristic points Ri, utilize Stationary Wavelet Transform method
Calculate the corresponding spatial coherence ρ (r of each characteristic pointi);
Threshold value E=0.05 is arranged in (2c), if obtained ρ (ri) meet ρ (ri) >=E, then by reference picture characteristic point riIt protects
It stays, otherwise, deletes this feature point;
(2d) traverses all characteristic points of reference picture, repeats step (2b)-(2c), the reference picture feature after being screened
Point;
(2e) calculates separately the Euclidean distance in reference picture set of characteristic points between any two characteristic point after above-mentioned screening
EdIf Ed>=15, then retain the two characteristic points, otherwise removes;
Preceding 10 characteristic points in the set of characteristic points that (2f) obtains step (2e) are as final reference picture feature
Point;
(3) multiscale image block feature construction reference picture feature point description is utilized to accord with:
(3a) arbitrarily chooses a reference picture characteristic point a, takes the image block P (a) of this feature point surrounding neighbors 15 × 15;
(3b) carries out multi-resolution decomposition using Stationary Wavelet Transform to image block P (a), obtains three different decomposition scales
Image block Ps(a), s=3,4,5;
(3c) calculates separately the grey level histogram vector H of above three different decomposition scale image blocks(a), as
The gray feature of feature descriptor;
(3d) calculates separately the gradient orientation histogram vector G of above three different decomposition scale image blocks(a), by it
Gradient Features as feature descriptor;
The corresponding gray feature of different decomposition scale image block and Gradient Features are together in series by (3e), obtain reference picture
The corresponding feature descriptor F (a) of characteristic point a={ H3(a),H4(a),H5(a),G3(a),G4(a),G5(a)};
(4) for any one image characteristic point b subject to registration, its characteristic point is obtained according to the same operation of step (3) and is retouched
Symbol F (b) is stated, the similitude between F (a) and image characteristic point descriptor F (b) subject to registration is accorded with using reference picture feature point description,
Establish the matching double points between reference picture and image subject to registration;
(5) abnormal point in the matching double points that removal step (4) obtains:
(5a) arbitrarily chooses a reference picture characteristic point r from the matching double points that step (4) obtainsc, matched
3 neighborhood points of this feature point arest neighbors are taken in reference picture characteristic point, and this 3 neighborhood points taken are mapped to and are referred to
Image characteristic point rcAs the image characteristic point t subject to registration of match pointcNeighborhood in, calculate reference picture characteristic point rcWith it
Geometry cost between neighborhood point
Wherein,Indicate reference picture characteristic point rcK-th of nearest neighbor point, tcIt indicates and reference picture characteristic point rcPhase
Matched image characteristic point subject to registration, m () indicate adaptation function, | | | | indicate that Euclidean distance, c indicate the rope of matching double points
Draw, value range is the index for the nearest neighbor point that 1 to 10, k indicates that c-th of reference picture characteristic point is taken, value range
It is 1 to 3;
(5b) traversal is all to have matched reference picture characteristic point, repeats step (5a), has been matched reference picture feature
Geometry cost between point and its respective neighborhood point, using the corresponding all characteristic points of geometry Least-cost value as benchmark point set
It closes, is expressed as:
Wherein, (rc,tc) indicate matching double points, tm(k)It indicates and neighborhood pointCorresponding match point;
(5c) calculates remaining match point to the geometry cost between datum mark using following formula:
Wherein, rc′And tc′Remaining reference picture characteristic point and characteristics of image subject to registration in matching double points are indicated respectively
Point,Indicate o-th of reference picture characteristic point in benchmark point set, tm(o)Indicate o-th of reference picture gather with collection on schedule in
The corresponding match point of characteristic point, the index of c ' expression residue matching double points, value range are the rope that 1 to 6, o indicates datum mark
Draw, value range is 1 to 4;
Threshold value E is arranged in (5d)o=0.03, if obtained geometry costMeetThen by (rc′,tc′) otherwise delete the matching double points as correct match point;
(5e) repeats step (5c)-(5d), obtains reference picture and the final matching double points of image subject to registration;
(6) according to final matching double points obtained above, affine Transform Model is established, calculates the geometric form of image subject to registration
Variable element, and the geometric deformation parameter is utilized, image subject to registration is subjected to geometric transformation, obtains registration result.
Compared with the prior art, the present invention has the following advantages:
First, since the present invention is during being registrated SAR image, characteristic point is selected using spatial coherence, simultaneously
Using multiscale image block feature construction feature descriptor, overcoming the prior art can not only with single scale image block message
The deficiency of accurate description characteristic point attribute so that the present invention improves the conspicuousness of characteristic point, enhances the Shandong to speckle noise
Stick.
Second, since the present invention establishes matching double points using the minimum difference criterion based on rarefaction representation technology, adopt simultaneously
Feature abnormalities point is filtered out with the geometrical-restriction relation between match point and its neighborhood point, the prior art is overcome and uses Euclidean distance ratio
Method is susceptible to the deficiency of error matching points during establishing matching double points so that the present invention improves the standard of match point
True property.
Description of the drawings
Fig. 1 is the implementation flow chart of the present invention;
Fig. 2 is first group of the simulation experiment result figure of the present invention;
Fig. 3 is second group of the simulation experiment result figure of the present invention.
Specific implementation mode
The present invention is described further below in conjunction with the accompanying drawings:
Referring to Fig.1, implementation steps of the invention are as follows:
Step 1, two images are inputted, an optional width, which is used as, refers to image I1, using another width as image I subject to registration2。
The two images of input are in two width SAR image of the different polarization modes obtained in certain airborne radar or different phases
It intercepts respectively.
Step 2, reference picture characteristic point is chosen.
2.1) SIFT algorithms extraction reference picture I is used1Characteristic point, by I1All characteristic points be stored in first set R
In;
2.2) a characteristic point r is arbitrarily chosen from reference picture set of characteristic points Ri, utilize Stationary Wavelet Transform method
Calculate the corresponding spatial coherence ρ (r of each characteristic pointi);
S Scale Decomposition 2.2a) is carried out to input picture using Stationary Wavelet Transform, obtains input picture in different scale
Upper 3 kinds of different detail pictures;
2.2b) define amplitude Ms of the arbitrary image pixel x under s scaless(x) it is expressed as:
Wherein,It is illustrated respectively in s scales hypograph in the horizontal direction, vertical direction and diagonal line side
Upward detailed information, | | indicate absolute value operation operation;
Following formula 2.2c) is used to calculate the spatial coherence ρ (x) of pixel x:
Wherein, ∏ () indicates multiplication operations;
2.3) threshold value E=0.05 is set, if obtained ρ (ri) meet ρ (ri) >=E, then by reference picture characteristic point riIt protects
It stays, otherwise, deletes this feature point;
2.4) all characteristic points of reference picture are traversed, step 2.2) -2.3 is repeated), the reference picture feature after being screened
Point;
2.5) Euclidean distance in reference picture set of characteristic points between any two characteristic point after above-mentioned screening is calculated separately
EdIf Ed>=15, then retain the two characteristic points, otherwise removes;
2.6) using preceding 10 characteristic points in the set of characteristic points that step 2.5) obtains as final reference picture feature
Point.
Step 3:It is accorded with using multiscale image block feature construction reference picture feature point description.
3.1) a reference picture characteristic point a is arbitrarily chosen, the image block P (a) of this feature point surrounding neighbors 15 × 15 is taken;
3.2) it uses Stationary Wavelet Transform to carry out multi-resolution decomposition to image block P (a), obtains three different decomposition scales
Image block Ps(a), s=3,4,5;
3.3) the grey level histogram vector H of above three different decomposition scale image block is calculated separatelys(a), as
The gray feature of feature point description symbol;
3.4) the gradient orientation histogram vector G of above three different decomposition scale image block is calculated separatelys(a), by it
Gradient Features as feature point description symbol;
3.5) the corresponding gray feature of different decomposition scale image block and Gradient Features are together in series, obtain reference picture
The corresponding feature descriptor F (a) of characteristic point a={ H3(a),H4(a),H5(a),G3(a),G4(a),G5(a)}。
Step 4:For any one image characteristic point b subject to registration, according to step (3), similarly operation obtains its characteristic point
Descriptor F (b), it is similar between image characteristic point descriptor F (b) subject to registration using reference picture feature point description symbol F (a)
Property, establish the matching double points between reference picture and image subject to registration.
There are many kinds of the methods for forming matching double points, common are arest neighbors method, Euclidean distance ratio etc., is adopted in this example
With but be not limited to the minimum difference criterion based on rarefaction representation technology and establish matching double points, specific implementation step is as follows:
4.1) a characteristic point r is arbitrarily chosen from the reference picture characteristic point that step (2) obtainsi, it is corresponding to calculate its
Feature descriptor F (ri);
4.2) the maximum offset between reference picture and image subject to registration is set as l=100, is chosen in image subject to registration
The window W of L × L sizesiAs characteristic point riRegion of search, and by all pixels point { V in the regionq:Q=1,2 ..., Q, Q
=L × L } it is used as characteristic point riCandidate matches point, wherein L=2 × l+1;
4.3) for above-mentioned each candidate matches point Vq, the window of the pixel surrounding neighbors 20 × 20 is taken, and will
All pixels point u in the windownCorresponding feature descriptor F (un) it is used as VqCorresponding sparse dictionary Dq={ F (un), n=1,
2 ..., J, J=20 × 20 };
4.4) it utilizes orthogonal matching pursuit algorithm to solve following formula, obtains characteristic point riIn sparse dictionary DqUnder it is sparse
Vectorial αq
Wherein, the value of independent variable when argmin () representative function reaches minimum value, | | | | indicate Euclidean distance, |
|·||0Indicate that zero norm, C indicate degree of rarefication;
4.5) cycling among windows WiInterior all pixels point repeats step 4.3) -4.4), obtain characteristic point riIn different dilute of Q
Dredge the sparse vector α under dictionaryq;
4.6) characteristic point r is calculated as followsiWith arbitrary candidate matches point VqBetween reconstructed error, and will have minimal reconstruction
The pixel of error is as characteristic point riMatch point m (ri)
4.7) all reference picture characteristic points are traversed, step 4.1) -4.6 is repeated), it obtains all reference picture characteristic points and exists
All characteristic points of image subject to registration are stored in second set T by corresponding match point in image subject to registration.
Step 5:Abnormal point in the matching double points that removal step (4) obtains.
5.1) a reference picture characteristic point r is arbitrarily chosen from the matching double points that step (4) obtainsc, matched
3 neighborhood points of this feature point arest neighbors are taken in reference picture characteristic point, and this 3 neighborhood points taken are mapped to and are referred to
Image characteristic point rcAs the image characteristic point t subject to registration of match pointcNeighborhood in, calculate reference picture characteristic point rcWith it
Geometry cost between neighborhood point
Wherein,Indicate reference picture characteristic point rcK-th of nearest neighbor point, tcIt indicates and reference picture characteristic point rcPhase
Matched image characteristic point subject to registration, m () indicate adaptation function, | | | | indicate that Euclidean distance, c indicate the rope of matching double points
Draw, value range is the index for the nearest neighbor point that 1 to 10, k indicates that c-th of reference picture characteristic point is taken, value range
It is 1 to 3;
5.2) traversal is all has matched reference picture characteristic point, repeats step 5.1), has been matched reference picture feature
Geometry cost between point and its respective neighborhood point, using the corresponding all characteristic points of geometry Least-cost value as benchmark point set
It closes, is expressed as:
Wherein, (rc,tc) indicate matching double points, tm(k)It indicates and neighborhood pointCorresponding match point;
5.3) following formula is used to calculate remaining match point to the geometry cost between datum mark:
Wherein, rc′And tc′Remaining reference picture characteristic point and characteristics of image subject to registration in matching double points are indicated respectively
Point,Indicate o-th of reference picture characteristic point in benchmark point set, tm(o)Indicate o-th of reference picture gather with collection on schedule in
The corresponding match point of characteristic point, the index of c ' expression residue matching double points, value range are the rope that 1 to 6, o indicates datum mark
Draw, value range is 1 to 4;
5.4) setting threshold value Eo=0.03, if obtained geometry costMeetThen by (rc′,tc′) otherwise delete the matching double points as correct match point;
5.5) step 5.3) -5.4 is repeated), obtain reference picture and the final matching double points of image subject to registration.
Step 6:According to final matching double points, registration result is obtained.
Affine Transform Model is established according to final matching double points obtained above, calculates the geometric deformation ginseng of image subject to registration
Number, this example specifically uses least square method computational geometry deformation parameter, and utilizes the geometric deformation parameter, by image subject to registration
Geometric transformation is carried out, registration result is obtained.
Effect of the present invention is described further with reference to experiment simulation.
1. simulated conditions:
The Simulation Experimental Platform of the present invention uses Intel (R) Pentium (R) CPU G3240 3.10GHz, inside saves as
4GB runs the PC machine of Windows 7, programming language Matlab2011b.
2. emulation content and interpretation of result:
The method for registering images based on SIFT-OCT, the method for registering images based on BFSIFT, base are applied in emulation 1 respectively
SAR image is registrated in the method for registering images and the present invention of NDSS-SIFT, the results are shown in Figure 2, and wherein Fig. 2 (a) is
Match point line graph based on SIFT-OCT methods, Fig. 2 (b) are the match point line graphs based on BFSIFT methods, and Fig. 2 (c) is
Match point line graph based on NDSS-SIFT methods, Fig. 2 (d) are the match point line graphs of the present invention.Yellow solid line table in Fig. 2
Show that correct matching double points, red line indicate error matching points pair.
The method for registering images based on SIFT-OCT, the method for registering images based on BFSIFT, base are applied in emulation 2 respectively
SAR image is registrated in the method for registering images and the present invention of NDSS-SIFT, the results are shown in Figure 3, and wherein Fig. 3 (a) is
Match point line graph based on SIFT-OCT methods, Fig. 3 (b) are the match point line graphs based on BFSIFT methods, and Fig. 3 (c) is
Match point line graph based on NDSS-SIFT methods, Fig. 3 (d) are the match point line graphs of the present invention.Yellow solid line table in Fig. 3
Show that correct matching double points, red line indicate error matching points pair.
From Fig. 2 (a) -2 (c) and Fig. 3 (a) -3 (c) as can be seen that for 2 groups of actual measurement SAR images pair, SIFT-OCT methods,
All there are more error matching points in 3 kinds of algorithms of BFSIFT methods and NDSS-SIFT methods.Wherein, SIFT-OCT methods are imitated
Fruit is worst, including error matching points it is most.Error matching points in BFSIFT methods and NDSS-SIFT methods are to relatively
It is few.
From Fig. 2 (d) and Fig. 3 (d) as can be seen that for 2 groups of actual measurement SAR images pair, the registration result that the present invention obtains is more
Accurately, error matching points are not included.This chooses reliable characteristic point derived from the present invention using spatial coherence, while utilizing more rulers
Image block characteristics construction feature descriptor is spent, the conspicuousness of feature descriptor and the robustness to speckle noise are enhanced.In addition,
Matching double points are established using the minimum difference criterion based on rarefaction representation technology, the accuracy of matching double points are improved, to solve
The prior art of having determined is applied to SAR image and matches the problem for occurring a large amount of error matching points pair on time.
Claims (2)
1. a kind of SAR image registration method based on multi-scale image block feature and rarefaction representation, includes the following steps:
(1) two images are inputted, an optional width, which is used as, refers to image I1, using another width as image I subject to registration2;
(2) reference picture characteristic point is chosen:
(2a) is using SIFT algorithms extraction reference picture I1Characteristic point, and by I1All characteristic points be stored in first set R
In;
(2b) arbitrarily chooses a characteristic point r from reference picture set of characteristic points Ri, calculated using Stationary Wavelet Transform method every
The corresponding spatial coherence ρ (r of a characteristic pointi):
S Scale Decomposition 2b1) is carried out to input picture using Stationary Wavelet Transform, obtains input picture 3 kinds on different scale
Different detail pictures;
2b2) define amplitude Ms of the arbitrary image pixel x under s scaless(x) it is expressed as:
Wherein, Ws h, Ws v, Ws dIt is illustrated respectively in s scales hypograph in the horizontal direction, vertical direction and diagonally adjacent thin
Information is saved, | | indicate absolute value operation operation;
Following formula 2b3) is used to calculate the spatial coherence ρ (x) of pixel x:
Wherein, Π () indicates multiplication operations;
Threshold value E=0.05 is arranged in (2c), if obtained ρ (ri) meet ρ (ri) >=E, then by reference picture characteristic point riRetain, it is no
Then, this feature point is deleted;
(2d) traverses all characteristic points of reference picture, repeats step (2b)-(2c), the reference picture characteristic point after being screened;
(2e) calculates separately the Euclidean distance E in reference picture set of characteristic points between any two characteristic point after above-mentioned screeningdIf
Ed>=15, then retain the two characteristic points, otherwise removes;
Preceding 10 characteristic points in the set of characteristic points that (2f) obtains step (2e) are as final reference picture characteristic point;
(3) multiscale image block feature construction reference picture feature point description is utilized to accord with:
(3a) arbitrarily chooses a reference picture characteristic point a, takes the image block P (a) of this feature point surrounding neighbors 15 × 15;
(3b) carries out multi-resolution decomposition using Stationary Wavelet Transform to image block P (a), obtains the image of three different decomposition scales
Block Ps(a), s=3,4,5;
(3c) calculates separately the grey level histogram vector H of above three different decomposition scale image blocks(a), it is retouched as feature
State the gray feature of symbol;
(3d) calculates separately the gradient orientation histogram vector G of above three different decomposition scale image blocks(a), as spy
Levy the Gradient Features of descriptor;
The corresponding gray feature of different decomposition scale image block and Gradient Features are together in series by (3e), obtain reference picture feature
The corresponding feature descriptor F (a) of point a={ H3(a),H4(a),H5(a),G3(a),G4(a),G5(a)};
(4) for any one image characteristic point b subject to registration, according to step (3), similarly operation obtains its feature point description symbol F
(b), using the similitude between reference picture feature point description symbol F (a) and image characteristic point descriptor F (b) subject to registration, ginseng is established
Examine the matching double points between image and image subject to registration;
(5) abnormal point in the matching double points that removal step (4) obtains:
(5a) arbitrarily chooses a reference picture characteristic point r from the matching double points that step (4) obtainsc, in matched reference chart
3 neighborhood points as taking this feature point arest neighbors in characteristic point, and this 3 neighborhood points taken are mapped to and reference picture spy
Levy point rcAs the image characteristic point t subject to registration of match pointcNeighborhood in, calculate reference picture characteristic point rcWith its neighborhood point
Between geometry cost
Wherein,Indicate reference picture characteristic point rcK-th of nearest neighbor point, tcIt indicates and reference picture characteristic point rcMatch
Image characteristic point subject to registration, m () indicate adaptation function, | | | | indicate Euclidean distance, c indicate matching double points index,
Its value range is the index for the nearest neighbor point that 1 to 10, k indicates that c-th of reference picture characteristic point is taken, value range 1
To 3;
(5b) traversal is all to have matched reference picture characteristic point, repeats step (5a), matched reference picture characteristic point with
Geometry cost between its respective neighborhood point, using the corresponding all characteristic points of geometry Least-cost value as benchmark point set, table
It is shown as:
Wherein, (rc,tc) indicate matching double points, tm(k)It indicates and neighborhood pointCorresponding match point;
(5c) calculates remaining match point to the geometry cost between datum mark using following formula:
Wherein, rc′And tc′Remaining reference picture characteristic point and image characteristic point subject to registration in matching double points are indicated respectively,
Indicate o-th of reference picture characteristic point in benchmark point set, tm(o)It indicates and o-th of reference picture feature in benchmark point set
The corresponding match point of point, the index of c ' expression residue matching double points, value range are the index that 1 to 6, o indicates datum mark,
Value range is 1 to 4;
Threshold value E is arranged in (5d)o=0.03, if obtained geometry costMeetThen
By (rc′,tc′) otherwise delete the matching double points as correct match point;
(5e) repeats step (5c)-(5d), obtains reference picture and the final matching double points of image subject to registration;
(6) according to final matching double points obtained above, affine Transform Model is established, calculates the geometric deformation ginseng of image subject to registration
Number, and the geometric deformation parameter is utilized, image subject to registration is subjected to geometric transformation, obtains registration result.
2. the SAR image registration method according to claim 1 based on multi-scale image block feature and rarefaction representation, special
Sign is to establish the matching double points between reference picture and image subject to registration in the step (4), carry out as follows:
A characteristic point r 4a) is arbitrarily chosen from the reference picture characteristic point that step (2) obtainsi, calculate its corresponding feature and retouch
State symbol F (ri);
The maximum offset between reference picture and image subject to registration 4b) is set as l=100, L × L is chosen in image subject to registration
The window W of sizeiAs characteristic point riRegion of search, by all pixels point { V in the regionq:Q=1,2 ..., Q, Q=L ×
L } it is used as characteristic point riCandidate matches point, wherein L=2 × l+1;
4c) for above-mentioned each candidate matches point Vq, take the window of the pixel surrounding neighbors 20 × 20, and by the window
Interior all pixels point unCorresponding feature descriptor F (un) it is used as VqCorresponding sparse dictionary Dq={ F (un), n=1,2 ...,
J, J=20 × 20 };
4d) orthogonal matching pursuit algorithm is utilized to solve following formula, obtains characteristic point riIn sparse dictionary DqUnder sparse vector αq
Wherein, the value of independent variable when argmin () representative function reaches minimum value, | | | | indicate Euclidean distance, | | | |0
Indicate that zero norm, C indicate degree of rarefication;
4e) cycling among windows WiInterior all pixels point repeats step 4c) -4d), obtain characteristic point riIn Q different sparse dictionaries
Under sparse vector αq;
Characteristic point r 4f) is calculated as followsiWith arbitrary candidate matches point VqBetween reconstructed error, and will be with minimal reconstruction error
Pixel is as characteristic point riMatch point m (ri)
All reference picture characteristic points 4g) are traversed, step 4a is repeated) -4f), all reference picture characteristic points are obtained subject to registration
All characteristic points of image subject to registration are stored in second set T by corresponding match point in image.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610118533.9A CN105787943B (en) | 2016-03-03 | 2016-03-03 | SAR image registration method based on multi-scale image block feature and rarefaction representation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610118533.9A CN105787943B (en) | 2016-03-03 | 2016-03-03 | SAR image registration method based on multi-scale image block feature and rarefaction representation |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105787943A CN105787943A (en) | 2016-07-20 |
CN105787943B true CN105787943B (en) | 2018-08-31 |
Family
ID=56386825
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610118533.9A Active CN105787943B (en) | 2016-03-03 | 2016-03-03 | SAR image registration method based on multi-scale image block feature and rarefaction representation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105787943B (en) |
Families Citing this family (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108627835B (en) * | 2018-06-29 | 2021-07-27 | 中国科学院电子学研究所 | Target reconstruction method of fully-polarized differential SAR (synthetic aperture radar) chromatography |
CN109166336B (en) * | 2018-10-19 | 2020-08-07 | 福建工程学院 | Real-time road condition information acquisition and pushing method based on block chain technology |
CN109741275B (en) * | 2018-12-28 | 2020-06-12 | 济南大学 | MVCT image enhancement method and system |
CN109816619B (en) * | 2019-01-28 | 2024-06-21 | 努比亚技术有限公司 | Image fusion method, device, terminal and computer readable storage medium |
CN110033422B (en) * | 2019-04-10 | 2021-03-23 | 北京科技大学 | Fundus OCT image fusion method and device |
CN112184785B (en) * | 2020-09-30 | 2023-03-24 | 西安电子科技大学 | Multi-mode remote sensing image registration method based on MCD measurement and VTM |
CN116008911B (en) * | 2022-12-02 | 2023-08-22 | 南昌工程学院 | Orthogonal matching pursuit sound source identification method based on novel atomic matching criteria |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103177444A (en) * | 2013-03-08 | 2013-06-26 | 中国电子科技集团公司第十四研究所 | Automatic SAR (synthetic-aperture radar) image rectification method |
CN103839265A (en) * | 2014-02-26 | 2014-06-04 | 西安电子科技大学 | SAR image registration method based on SIFT and normalized mutual information |
-
2016
- 2016-03-03 CN CN201610118533.9A patent/CN105787943B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103177444A (en) * | 2013-03-08 | 2013-06-26 | 中国电子科技集团公司第十四研究所 | Automatic SAR (synthetic-aperture radar) image rectification method |
CN103839265A (en) * | 2014-02-26 | 2014-06-04 | 西安电子科技大学 | SAR image registration method based on SIFT and normalized mutual information |
Non-Patent Citations (2)
Title |
---|
Change detection based on image segment and fusion in multitemporal SAR images;Guiting Wang等;《Synthetic Aperture Radar, 2009. APSAR 2009. 2nd Asian-Pacific Conference on》;20100108;第821-824页 * |
基于MAR-MRF的SAR图像分割方法;刘爱平等;《电子与信息学报》;20091130;第31卷(第11期);第2557-2562页 * |
Also Published As
Publication number | Publication date |
---|---|
CN105787943A (en) | 2016-07-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105787943B (en) | SAR image registration method based on multi-scale image block feature and rarefaction representation | |
US9824294B2 (en) | Saliency information acquisition device and saliency information acquisition method | |
JP2019514123A (en) | Remote determination of the quantity stored in containers in geographical areas | |
JP6330385B2 (en) | Image processing apparatus, image processing method, and program | |
Sidorov | Conditional gans for multi-illuminant color constancy: Revolution or yet another approach? | |
CN103886611B (en) | A kind of image matching method for being suitable for aeroplane photography flight reappearance and checking automatically | |
CN104200461B (en) | The remote sensing image registration method of block and sift features is selected based on mutual information image | |
CN104751465A (en) | ORB (oriented brief) image feature registration method based on LK (Lucas-Kanade) optical flow constraint | |
CN103914847A (en) | SAR image registration method based on phase congruency and SIFT | |
Fytsilis et al. | A methodology for near real-time change detection between Unmanned Aerial Vehicle and wide area satellite images | |
CN104504723A (en) | Image registration method based on remarkable visual features | |
CN105335952B (en) | Matching power flow computational methods and device and parallax value calculating method and equipment | |
CN107862680B (en) | Target tracking optimization method based on correlation filter | |
CN107516322A (en) | A kind of image object size based on logarithm pole space and rotation estimation computational methods | |
CN108550165A (en) | A kind of image matching method based on local invariant feature | |
CN103077528A (en) | Rapid image matching method based on DCCD (Digital Current Coupling)-Laplace and SIFT (Scale Invariant Feature Transform) descriptors | |
CN103500454A (en) | Method for extracting moving target of shaking video | |
CN113763274A (en) | Multi-source image matching method combining local phase sharpness orientation description | |
CN111126508A (en) | Hopc-based improved heterogeneous image matching method | |
CN111563423A (en) | Unmanned aerial vehicle image target detection method and system based on depth denoising automatic encoder | |
Sharma et al. | Study and analysis of edge detection techniques in digital images | |
CN111274858A (en) | Business license identification method in network transaction supervision | |
Salehpour et al. | 3D face reconstruction by KLT feature extraction and model consistency match refining and growing | |
Phogat et al. | Different image registration methods—an overview | |
Csillik | Superpixels: The end of pixels in obia. A comparison of stat-of-the-art superpixel methods for remote sensing data |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |