CN106296611B - A kind of compressed sensing image recovery method and its system using objective attribute target attribute auxiliary - Google Patents
A kind of compressed sensing image recovery method and its system using objective attribute target attribute auxiliary Download PDFInfo
- Publication number
- CN106296611B CN106296611B CN201610647440.5A CN201610647440A CN106296611B CN 106296611 B CN106296611 B CN 106296611B CN 201610647440 A CN201610647440 A CN 201610647440A CN 106296611 B CN106296611 B CN 106296611B
- Authority
- CN
- China
- Prior art keywords
- target
- subspace
- dictionary
- vector
- infrared small
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
- 238000000034 method Methods 0.000 title claims abstract description 24
- 238000011084 recovery Methods 0.000 title claims abstract description 21
- 238000005192 partition Methods 0.000 claims abstract description 16
- 239000011159 matrix material Substances 0.000 claims description 59
- 230000008901 benefit Effects 0.000 claims description 11
- 238000005516 engineering process Methods 0.000 claims description 9
- 230000006835 compression Effects 0.000 description 6
- 238000007906 compression Methods 0.000 description 6
- 238000002474 experimental method Methods 0.000 description 5
- 230000008859 change Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000007152 ring opening metathesis polymerisation reaction Methods 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000005057 refrigeration Methods 0.000 description 2
- 238000003860 storage Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
-
- 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
- G06T2207/10004—Still image; Photographic image
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Aiming, Guidance, Guns With A Light Source, Armor, Camouflage, And Targets (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of compressed sensing image recovery method assisted using objective attribute target attribute, wherein the described method includes: initialization step, Subspace partition step, atom collection update step, sparse coefficient updates step, output step.The present invention also provides a kind of compressed sensing image recovery systems assisted using objective attribute target attribute.Technical solution provided by the invention is in the case where Small object picture signal degree of rarefication is unknown, it will reflect that the auxiliary information of target signature is introduced into the division of subspace, to accurately and effectively select dictionary subspace the most matched, the efficient and quick reconstruct of Small object picture signal is realized.
Description
Technical field
The present invention relates to image procossing compressed sensing field more particularly to a kind of compressed sensings assisted using objective attribute target attribute
Image recovery method and its system.
Background technique
Compressed sensing technology is widely used in image procossing neck because being expected to the balance of realization compression ratio and reconstruction quality
Domain, and be forward position and the hot spot of signal processing neighborhood since its birth.Image is as a kind of natural sign, between adjacent pixel
Correlation caused by spatial redundancy, there are time redundancy caused by correlation, different color planes or frequency spectrums between different frame
Spectral redundancy caused by the correlation of band, i.e., there is redundancies in image data, and image has closely on some sparse domains
Like sparsity, it is possible to carry out compression and reconstruction processing to image using compressive sensing theory.However, existing restructing algorithm
In fail in reconstruction quality into obtaining preferable balance between reconstructed velocity, they often focus on reconstruction quality and have ignored weight
The structure time, in the case, there has been proposed stator spatial class recovery algorithms.
Subspace class OMP algorithm because having the advantages of simple structure and easy realization, and it is lower compared to other types algorithm operation quantity and by
Extensive utilization is into actual engineering project.Such algorithm includes OMP, StOMP, ROMP, CoSaMP and SP algorithm.OMP algorithm
The size of its subspace is 1, therefore will lead to when degree of rarefication K is excessive that the number of iterations is more, and operand is big, and reconstructed velocity is slow.
Multiple matched atoms may be selected in StOMP algorithm an iteration process, so the number of iterations is effectively reduced, and then reduce the fortune of algorithm
Calculation amount reduces reconstitution time.But it is very strong to the dependence of degree of rarefication, only correctly estimating degree of rarefication K could correctly reconstruct
Signal out.ROMP algorithm is to be screened twice according to relevance principle and regularization process to atom, but the algorithm neutron
The selection in space is relied on dependent on correlation between value, and to degree of rarefication K transition, and the correctness of degree of rarefication estimation will affect algorithm
Convergence rate, quality reconstruction etc..CoSaMP algorithm selects multiple uncorrelated atoms simultaneously by the thought of backtracking from atom
The uncorrelated atom in part is rejected, multiple matched atoms may be selected in an iteration.But it equally needs known degree of rarefication K, however
True signal degree of rarefication K is often unknown, secondly, increasing and rejecting based on atom when the every step iteration of CoSaMP algorithm
Principle is different, and the difference of standard may result in the inaccuracy of the estimation to supported collection.SP algorithm is the compromise of above-mentioned algorithm,
Performance is optimal in above-mentioned algorithm.SP algorithm computation complexity is low, reconstruction accuracy is high, and has stringent theoretical guarantee, when
When observing matrix A meets limited equidistant property, SP algorithm can go out the sparse letter of any K- by Accurate Reconstruction from the observation of its noiseless
Number.It has similar advantage and disadvantage with CoSaMP algorithm.
Although above-mentioned all algorithms have been all made of the concept of fixed subspace, in addition to the subspace size of OMP algorithm is 1,
Remaining algorithm subspace size is all larger than 1, and these algorithms make the number of iterations tail off because once selecting multiple matched atoms,
And then reduce operand and reconstitution time.But still fail to the requirement for reaching real-time, and above-mentioned algorithm depend on unduly it is dilute
The estimation correctness for dredging degree K, K will determine that can signal be estimated correctly the time with signal reconstruction, meanwhile, in these algorithms
The selection of dictionary subspace be to be completed according to correlation between value, do not consider the physical feature attribute of picture signal.
Summary of the invention
In view of this, the purpose of the present invention is to provide a kind of compressed sensing image recovery sides assisted using objective attribute target attribute
Method and its system, it is intended to which how solution in the case where Small object picture signal degree of rarefication is unknown, passes through target in the prior art
Characteristic attribute (such as target size apriority, coherency and airspace sparsity etc.) come to existing for sub- spatial class OMP class algorithm
The problem of the defects of reconstructed velocity is lower compared with slow, reconstruction accuracy and depends on degree of rarefication unduly is made up.
The present invention proposes a kind of compressed sensing image recovery method assisted using objective attribute target attribute, which is characterized in that described
Method includes:
Initialization step: by redundancy vector rt-1, atom collection A and the number of iterations t these parameters are initialized, wherein
rt-1Decrement direction finding the amount y, A for being initialized as input are initialized as empty set, and t is initialized as 1；
Subspace partition step: it generates and infrared small target image sequential matrix D ∈ R of the same sizew×h, wherein w is
The length of infrared small target image, h are the width of infrared small target image, and 1 is saved in the sequential matrix D to w × h's
Number, while the sequential matrix D is divided by several two-dimentional sub-blocks according to preset target size, and by the sequential matrix D
Middle divided each sub-block is converted into column vector all in accordance with the mode that column stack, and is index with the element in each column vector
Corresponding vector is picked out from the Φ of dictionary space constitutes dictionary subspace Φi, i=1,2 ..., k, to realize to dictionary sky
Between division；
Atom collection updates step: finding out redundancy vector rt-1In each dictionary subspace Φ of divisioniIn projectionAnd calculate PiEnergyThe corresponding dictionary subspace of ceiling capacity is found out simultaneously
Index subscriptAccording to calculated index subscript λtFind out corresponding dictionary subspaceAnd benefit
Atom collection A is updated with dictionary subspace, i.e.,Wherein, AtIt is the atom collection currently updated, At-1It is
The atom collection that last iteration updates, after updating atom collection A each time, by λtCorresponding dictionary subspace0 is set to ensure
Do not repeat value；
Sparse coefficient updates step: utilizing the updated atom collection A of current iterationtLeast square is carried out to sparse coefficient to estimate
Meter, finds out signal in atom collection AtIn sparse coefficient componentAnd update redundancy vector rt=y-
Axt；
Export step: according to update redundancy vector r obtainedtIt is iterated stop technology, as redundancy vector rtNorm
When greater than preset constant ε, then jumps to the atom collection and update step progress next iteration, otherwise terminate iteration and export letter
The final estimated value of number sparse coefficient.
Preferably, the Subspace partition step specifically includes:
It is assumed that target size is B=a × b, then template size is set as a × b, generate corresponding infrared small target image size
Two-dimentional sequential matrix D ∈ Rw×h, wherein a, b are respectively target priori length and target priori width, and w, h are respectively infrared small
The length and width of target image, w divide exactly a, and h divides exactly b；
Element in two-dimentional sequential matrix D is the column stacking value of corresponding infrared small target image pixel point position, according to mesh
The priori size of dimensioning divides two-dimentional sequential matrix D, obtains D={ B1,B2,...,Bi,...Bk, wherein
As the aliquant target priori length a of infrared small target image length w, infrared small target picture traverse h cannot be whole
When except target priori width b, then according to the actual situation template size is set as that target can be completely covered, it also can be infrared
The value that the length and width of Small object image divide exactly；
The mode that each sub-block divided in the sequential matrix D stacks all in accordance with column is converted into column vector, with every
Element in one column vector is that index picks out corresponding vector composition dictionary subspace Φ from the Φ of dictionary spacei, i=1,
2 ..., k, i.e. division dictionary subspace, Φi=Φ (Vec (Bi)), i=1,2 ..., k.
On the other hand, the present invention also provides a kind of compressed sensing infrared small target image restorers of objective attribute target attribute auxiliary
System, the system comprises:
Initialization module is used for redundancy vector rt-1, atom collection A and the number of iterations t these parameters initialized,
Wherein, rt-1Decrement direction finding the amount y, A for being initialized as input are initialized as empty set, and t is initialized as 1；
Subspace partition module, for generating and infrared small target image sequential matrix D ∈ R of the same sizew×h, wherein
W is the length of infrared small target image, and h is the width of infrared small target image, and 1 to w × h is saved in the sequential matrix D
Number, while the sequential matrix D is divided by several two-dimentional sub-blocks according to preset target size, and by the sequential matrix
The each sub-block divided in D is converted into column vector all in accordance with the mode that column stack, using the element in each column vector as rope
Draw and picks out corresponding vector composition dictionary subspace Φ from the Φ of dictionary spacei, i=1,2 ..., k, to realize to dictionary
The division in space；
Atom collection update module, for finding out redundancy vector rt-1In each dictionary subspace Φ of divisioniIn projectionAnd calculate PiEnergyThe corresponding dictionary subspace of ceiling capacity is found out simultaneously
Index subscriptAccording to calculated index subscript λtFind out corresponding dictionary subspaceAnd it utilizes
Dictionary subspace is updated atom collection A, i.e.,Wherein, AtIt is the atom collection currently updated, At-1On being
The atom collection that an iteration updates, after updating atom collection A each time, by λtCorresponding dictionary subspace0 is set to ensure not
Repeat value；
Sparse coefficient update module, for utilizing the updated atom collection A of current iterationtMinimum two is carried out to sparse coefficient
Multiply estimation, finds out signal in atom collection AtIn sparse coefficient componentAnd update redundancy vector rt=
y-Axt；
Output module, for according to update redundancy vector r obtainedtIt is iterated stop technology, as redundancy vector rt's
It when norm is greater than preset constant ε, then jumps to the atom collection and updates step and carry out next iteration, otherwise terminate iteration and defeated
The final estimated value of signal sparse coefficient out.
Preferably, the Subspace partition module is specifically used for:
It is assumed that target size is B=a × b, then template size is set as a × b, generate corresponding infrared small target image size
Two-dimentional sequential matrix D ∈ Rw×h, wherein a, b are respectively target priori length and target priori width, and w, h are respectively infrared small
The length and width of target image, w divide exactly a, and h divides exactly b；
Element in two-dimentional sequential matrix D is the column stacking value of corresponding infrared small target image pixel point position, according to mesh
The priori size of dimensioning divides two-dimentional sequential matrix D, obtains D={ B1,B2,...,Bi,...Bk, wherein
As the aliquant target priori length a of infrared small target image length w, infrared small target picture traverse h cannot be whole
When except target priori width b, then according to the actual situation template size is set as that target can be completely covered, it also can be infrared
The value that the length and width of Small object image divide exactly；
The mode that each sub-block divided in the sequential matrix D stacks all in accordance with column is converted into column vector, with every
Element in one column vector is that index picks out corresponding vector composition dictionary subspace Φ from the Φ of dictionary spacei, i=1,
2 ..., k, i.e. division dictionary subspace, Φi=Φ (Vec (Bi)), i=1,2 ..., k.
Technical solution provided by the invention will reflect target spy in the case where Small object picture signal degree of rarefication is unknown
The auxiliary information of sign is introduced into the division of subspace, to accurately and effectively select dictionary subspace the most matched, is realized
The efficient and quick reconstruct of Small object picture signal.Compared with prior art, present invention has an advantage that (1) is of the invention not
The degree of rarefication for relying on signal eliminates the reliance on correlation selection dictionary subspace between value；(2) present invention can effectively realize target
Reconstruct, improve reconstruction quality；(3) present invention can effectively reduce the number of iterations and reduce operand, greatly improve
The speed of reconstruct.
Detailed description of the invention
Fig. 1 is the compressed sensing infrared small target image recovery method stream of objective attribute target attribute auxiliary in an embodiment of the present invention
Cheng Tu；
Fig. 2 is in an embodiment of the present invention using by refrigeration mode thermal infrared imager that model IR300, frame frequency are 50 frames
The foreground image of the infrared small target image acquired from different scenes --- single infrared small target image, two infrared small targets
Image and multiple infrared small target images；
Fig. 3 is the quality reconstruction figure in an embodiment of the present invention according to experiment condition setting to infrared small target image；
Fig. 4 is the inside of the compressed sensing image recovery system 10 assisted in an embodiment of the present invention using objective attribute target attribute
Structural schematic diagram.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.
Of the existing technology in order to solve the problems, such as, thinking proposed by the present invention is: carrying out to infrared small target image
It when reconstruct, is divided by objective attribute target attribute auxuliary subspace, and then assists infrared small target image reconstruction.The present invention is for existing solid
Degree of rarefication is relied in stator space class OMP algorithm and according to the problem of Correlation selection subspace, proposing according to target ruler between value
It is very little to divide dictionary subspace, it effectively reduces the number of iteration, promote reconstruction quality and independent of degree of rarefication.The present invention couple
The specific method of the reconstruct of infrared small target image: first generate with corresponding infrared small target image sequential matrix of the same size,
According to target priori size stripe sequence matrix, each piece in the sequential matrix of division is converted into according to the form that column stack
Column vector carries out selecting for dictionary subspace, referred to as stroke of subspace so that the element in each column vector is index set
Point.After dividing subspace, projection of the redundancy in each sub-spaces is found out with certain relevance principle, and find out maximal projection
It is worth corresponding dictionary subspace, atom set is expanded with the dictionary subspace, and then carry out the estimation of signal and redundancy.
A kind of compressed sensing image recovery method assisted using objective attribute target attribute provided by the invention, which is had, is not depending on letter
In the case where number degree of rarefication, the advantages of capable of efficiently and fast realizing the reconstruct of Small object picture signal, in practical applications can
Data volume, carrying cost and transmission bandwidth are enough substantially reduced, the time is saved.
It below will be to a kind of compressed sensing image recovery method progress assisted using objective attribute target attribute provided by the present invention
It is described in detail.
Referring to Fig. 1, the compressed sensing infrared small target image for objective attribute target attribute auxiliary in an embodiment of the present invention is extensive
Multiple method flow diagram.
In step sl, by redundancy vector rt-1, atom collection A and the number of iterations t these parameters are initialized, wherein
rt-1Decrement direction finding the amount y, A for being initialized as input are initialized as empty set, and t is initialized as 1.
In step s 2, it Subspace partition step: generates and infrared small target image sequential matrix D ∈ R of the same sizew ×h, wherein w is the length of infrared small target image, and h is the width of infrared small target image, is saved in the sequential matrix D
1 to w × h number, while the sequential matrix D is divided by several two-dimentional sub-blocks according to preset target size, and will be described
The each sub-block divided in sequential matrix D is converted into column vector all in accordance with the mode that column stack, in each column vector
Element is that index picks out corresponding vector composition dictionary subspace Φ from the Φ of dictionary spacei, i=1,2 ..., k, thus real
Now to the division in dictionary space.
In the present embodiment, the Subspace partition step S2 is specifically included:
It is assumed that target size is B=a × b, then template size is set as a × b, generate corresponding infrared small target image size
Two-dimentional sequential matrix D ∈ Rw×h, wherein a, b are respectively target priori length and target priori width, and w, h are respectively infrared small
The length and width of target image, w divide exactly a, and h divides exactly b；
Element in two-dimentional sequential matrix D is the column stacking value of corresponding infrared small target image pixel point position, according to mesh
The priori size of dimensioning divides two-dimentional sequential matrix D, obtains D={ B1,B2,...,Bi,...Bk, wherein
As the aliquant target priori length a of infrared small target image length w, infrared small target picture traverse h cannot be whole
When except target priori width b, then according to the actual situation template size is set as that target can be completely covered, it also can be infrared
The value that the length and width of Small object image divide exactly；
The mode that each sub-block divided in the sequential matrix D stacks all in accordance with column is converted into column vector, with every
Element in one column vector is that index picks out corresponding vector composition dictionary subspace Φ from the Φ of dictionary spacei, i=1,
2 ..., k, i.e. division dictionary subspace, Φi=Φ (Vec (Bi)), i=1,2 ..., k.
In step s3, atom collection updates step: finding out redundancy vector rt-1In each dictionary subspace Φ of divisioni
In projectionAnd calculate PiEnergyThe corresponding word of ceiling capacity is found out simultaneously
The index subscript of allusion quotation subspaceAccording to calculated index subscript λtFind out corresponding dictionary subspaceAnd atom collection A is updated using dictionary subspace, i.e.,Wherein, AtIt is the atom currently updated
Collection, At-1It is the atom collection that last iteration updates, after updating atom collection A each time, by λtCorresponding dictionary subspaceIt sets
0 to ensure not repeat value.
In step s 4, sparse coefficient updates step: utilizing the updated atom collection A of current iterationtTo sparse coefficient into
Row least-squares estimation finds out signal in atom collection AtIn sparse coefficient componentAnd update redundancy
Vector rt=y-Axt。
In step s 5, step is exported: according to update redundancy vector r obtainedtIt is iterated stop technology, works as redundancy
Vector rtNorm when being greater than preset constant ε, then jump to the atom collection and update step and carry out next iteration, otherwise terminate
The final estimated value of iteration and output signal sparse coefficient.
Below in conjunction with specific embodiment, it is reconstructed to verify the method for the present invention to infrared small target picture signal
High efficiency and rapidity.The present invention is utilized by model IR300, and frame frequency is the refrigeration mode thermal infrared imager of 50 frames from different fields
The foreground image of the infrared small target image of scape acquisition --- single infrared small target image, two infrared small target images and more
A infrared small target image (as shown in Figure 2) is tested, and for aesthetics, finally merges the result of reconstruct with background to be formed
The original image of estimation.Reconstruction quality is wherein measured with more objective Y-PSNR, it is quick with reconstitution time measure algorithm
Property.
Experimental configuration of the invention specifically includes: two-dimentional sequential matrix D ∈ Rw×h, infrared small target image size is w × h,
W=240, h=320；Signal length N=w × h=76800 (the infrared small target picture signal dimension of column heap poststack)；Decrement
Survey dimension M=19200；Compression ratioPerceive matrix Φ ∈ RM×N(vector immediately of Gaussian distributed)；Compression
Measure y ∈ RM×1；(w divides exactly a, and h is divided exactly b) for target size B=a × b=4 × 4.
Two-dimentional sequential matrix D is divided according to the priori size of target size first, obtains D={ B1,B2,...,
Bi,...Bk, whereinSo corresponding Subspace partition is Φi=Φ (Vec (Bi)), i=1,
2,...,k.Projection P of the computing redundancy vector in each sub-spacesi, and find out PiENERGY Ei, while finding out ceiling capacity
The index subscript λ of corresponding dictionary subspacet.According to calculated index subscript λtFind out corresponding dictionary subspaceI.e.
With the strongest sub- dictionary of redundancy vector correlation, atom set is updated with the subspaceThe experiment
C=a × b=16 dictionary atom is just expanded by an iteration, i.e. one cycle can restore 16 sparse coefficient points.Pay attention to
After updating atom collection A each time, by λtCorresponding dictionary subspace0 is set to ensure not repeat value.
It is arranged according to above-mentioned experiment condition, the present invention is as shown in Figure 3 to the quality reconstruction figure of infrared small target image.
Above-mentioned experiment is only single compressed more next time than the quality reconstruction figure tested, to avoid influence caused by contingency,
And more objectively illustrate that the present invention is directed to the validity of infrared small target image reconstruction, it will be red to 100 under Same Scene
Outer Small object image is reconstructed, and counts the average peak signal to noise ratio that the present invention reconstructs under different compression ratios, the present invention
Average peak signal to noise ratio it is as shown in table 1.
The average peak signal to noise ratio of 1 the method for the present invention of table reconstruct
It being drawn a conclusion by Fig. 3 and table 1, the present invention is able to achieve reconstruct to the infrared small target image under different scenes, and
The Y-PSNR of reconstruct is higher, has absolutely proved effectiveness of the invention and its accuracy.
The present invention mainly solves the problems, such as that reconstitution time existing for stator spatial class algorithm is long, so will pass through herein
Experiment statistics go out under different compression ratios, and the present invention reconstructs the average time of different scenes infrared image, as shown in table 2.
The reconstitution time of 2 the method for the present invention of table
By table 2 it can be concluded that two o'clock conclusion, first is that the present invention can go out different scenes with accurate reconstruction when compressing relatively low
Infrared small target image；Second is that the present invention is smaller to the reconstitution time of infrared small target figure.Above-mentioned conclusion has absolutely proved this hair
Bright rapidity.
A kind of compressed sensing image recovery method assisted using objective attribute target attribute provided by the invention, is believed in Small object image
In the case that number degree of rarefication is unknown, it will reflect that the auxiliary information of target signature is introduced into the division of subspace, thus accurately
Effective selection dictionary subspace the most matched, realizes the efficient and quick reconstruct of Small object picture signal.With existing skill
Art is compared, present invention has an advantage that (1) present invention does not depend on the degree of rarefication of signal, eliminates the reliance on correlation selection word between value
Allusion quotation subspace；(2) present invention can effectively realize the reconstruct of target, improve reconstruction quality；(3) present invention can effectively subtract
Lack the number of iterations and reduce operand, greatly improves the speed of reconstruct.
The specific embodiment of the invention also provides a kind of compressed sensing image recovery system 10 assisted using objective attribute target attribute,
It specifically includes that
Initialization module 11 is used for redundancy vector rt-1, atom collection A and these parameters of the number of iterations t carry out it is initial
Change, wherein rt-1Decrement direction finding the amount y, A for being initialized as input are initialized as empty set, and t is initialized as 1；
Subspace partition module 12, for generating and infrared small target image sequential matrix D ∈ R of the same sizew×h,
In, w is the length of infrared small target image, and h is the width of infrared small target image, and 1 to w is saved in the sequential matrix D
The number of × h, while the sequential matrix D is divided by several two-dimentional sub-blocks according to preset target size, and by the sequence
The each sub-block divided in matrix D is converted into column vector all in accordance with the mode that column stack, with the element in each column vector
Corresponding vector composition dictionary subspace Φ is picked out from the Φ of dictionary space to indexi, i=1,2 ..., k, thus realization pair
The division in dictionary space；
Atom collection update module 13, for finding out redundancy vector rt-1In each dictionary subspace Φ of divisioniIn throwing
ShadowAnd calculate PiEnergyIt is empty that corresponding dictionary of ceiling capacity is found out simultaneously
Between index subscriptAccording to calculated index subscript λtFind out corresponding dictionary subspaceAnd
Atom collection A is updated using dictionary subspace, i.e.,Wherein, AtIt is the atom collection currently updated, At-1
It is the atom collection that last iteration updates, after updating atom collection A each time, by λtCorresponding dictionary subspace0 is set with true
It protects and does not repeat value；
Sparse coefficient update module 14, for utilizing the updated atom collection A of current iterationtSparse coefficient is carried out minimum
Two multiply estimation, find out signal in atom collection AtIn sparse coefficient componentAnd update redundancy vector rt
=y-Axt；
Output module 15, for according to update redundancy vector r obtainedtIt is iterated stop technology, as redundancy vector rt
Norm when being greater than preset constant ε, then jump to the atom collection and update step and carry out next iteration, otherwise terminate iteration simultaneously
The final estimated value of output signal sparse coefficient.
The compressed sensing infrared small target image recovery system 10 of a kind of objective attribute target attribute auxiliary provided by the invention, in small mesh
In the case that logo image signal degree of rarefication is unknown, it will reflect that the auxiliary information of target signature is introduced into the division of subspace,
To accurately and effectively select dictionary subspace the most matched, the efficient and quick reconstruct of Small object picture signal is realized.
Restored referring to Fig. 4, showing in an embodiment of the present invention using the compressed sensing image that objective attribute target attribute assists
The structural schematic diagram of system 10.
In the present embodiment, the compressed sensing image recovery system 10 assisted using objective attribute target attribute, main includes initial
Change module 11, Subspace partition module 12, atom collection update module 13, sparse coefficient update module 14 and output module 15.
Initialization module 11 is used for redundancy vector rt-1, atom collection A and these parameters of the number of iterations t carry out it is initial
Change, wherein rt-1Decrement direction finding the amount y, A for being initialized as input are initialized as empty set, and t is initialized as 1.
Subspace partition module 12, for generating and infrared small target image sequential matrix D ∈ R of the same sizew×h,
In, w is the length of infrared small target image, and h is the width of infrared small target image, and 1 to w is saved in the sequential matrix D
The number of × h, while the sequential matrix D is divided by several two-dimentional sub-blocks according to preset target size, and by the sequence
The each sub-block divided in matrix D is converted into column vector all in accordance with the mode that column stack, with the element in each column vector
Corresponding vector composition dictionary subspace Φ is picked out from the Φ of dictionary space to indexi, i=1,2 ..., k, thus realization pair
The division in dictionary space.
In the present embodiment, the Subspace partition module 12 is specifically used for:
It is assumed that target size is B=a × b, then template size is set as a × b, generate corresponding infrared small target image size
Two-dimentional sequential matrix D ∈ Rw×h, wherein a, b are respectively target priori length and target priori width, and w, h are respectively infrared small
The length and width of target image, w divide exactly a, and h divides exactly b；
Element in two-dimentional sequential matrix D is the column stacking value of corresponding infrared small target image pixel point position, according to mesh
The priori size of dimensioning divides two-dimentional sequential matrix D, obtains D={ B1,B2,...,Bi,...Bk, wherein
As the aliquant target priori length a of infrared small target image length w, infrared small target picture traverse h cannot be whole
When except target priori width b, then according to the actual situation template size is set as that target can be completely covered, it also can be infrared
The value that the length and width of Small object image divide exactly；
The mode that each sub-block divided in the sequential matrix D stacks all in accordance with column is converted into column vector, with every
Element in one column vector is that index picks out corresponding vector composition dictionary subspace Φ from the Φ of dictionary spacei, i=1,
2 ..., k, i.e. division dictionary subspace, Φi=Φ (Vec (Bi)), i=1,2 ..., k.
Atom collection update module 13, for finding out redundancy vector rt-1In each dictionary subspace Φ of divisioniIn throwing
ShadowAnd calculate PiEnergyIt is empty that corresponding dictionary of ceiling capacity is found out simultaneously
Between index subscriptAccording to calculated index subscript λtFind out corresponding dictionary subspaceAnd benefit
Atom collection A is updated with dictionary subspace, i.e.,Wherein, AtIt is the atom collection currently updated, At-1It is
The atom collection that last iteration updates, after updating atom collection A each time, by λtCorresponding dictionary subspace0 is set to ensure
Do not repeat value.
Sparse coefficient update module 14, for utilizing the updated atom collection A of current iterationtSparse coefficient is carried out minimum
Two multiply estimation, find out signal in atom collection AtIn sparse coefficient componentAnd update redundancy vector rt
=y-Axt。
Output module 15, for according to update redundancy vector r obtainedtIt is iterated stop technology, as redundancy vector rt
Norm when being greater than preset constant ε, then jump to the atom collection and update step and carry out next iteration, otherwise terminate iteration simultaneously
The final estimated value of output signal sparse coefficient.
The compressed sensing infrared small target image recovery system 10 of a kind of objective attribute target attribute auxiliary provided by the invention, in small mesh
In the case that logo image signal degree of rarefication is unknown, it will reflect that the auxiliary information of target signature is introduced into the division of subspace,
To accurately and effectively select dictionary subspace the most matched, the efficient and quick reconstruct of Small object picture signal is realized.
Compared with prior art, present invention has an advantage that (1) present invention does not depend on the degree of rarefication of signal, correlation between value is eliminated the reliance on
Property selection dictionary subspace；(2) present invention can effectively realize the reconstruct of target, improve reconstruction quality；(3) energy of the present invention
It enough effectively reduces the number of iterations and reduces operand, greatly improve the speed of reconstruct.
The present invention has in the case where not depending on signal degree of rarefication, can efficiently and fast realize that Small object image is believed
Number reconstruct the advantages of, data volume, carrying cost and transmission bandwidth can be substantially reduced in practical applications, save the time.
The present invention is widely used, suitable for the Small object image procossing of multiple application fields, such as the infrared figure of military field
Picture, the CT image of medical domain and Mura image of field of industrial production etc..
It is worth noting that, included each unit is only divided according to the functional logic in above-described embodiment,
But it is not limited to the above division, as long as corresponding functions can be realized；In addition, the specific name of each functional unit
It is only for convenience of distinguishing each other, the protection scope being not intended to restrict the invention.
In addition, those of ordinary skill in the art will appreciate that realizing all or part of the steps in the various embodiments described above method
It is that relevant hardware can be instructed to complete by program, corresponding program can store to be situated between in a computer-readable storage
In matter, the storage medium, such as ROM/RAM, disk or CD.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Made any modifications, equivalent replacements, and improvements etc., should all be included in the protection scope of the present invention within mind and principle.
Claims (4)
1. a kind of compressed sensing image recovery method assisted using objective attribute target attribute, which is characterized in that the described method includes:
Initialization step: by redundancy vector rt-1, atom collection A and the number of iterations t these parameters are initialized, wherein rt-1
Decrement direction finding the amount y, A for being initialized as input are initialized as empty set, and t is initialized as 1；
Subspace partition step: it generates and infrared small target image sequential matrix D ∈ R of the same sizew×h, wherein w is infrared
The length of Small object image, h are the width of infrared small target image, 1 to w × h number are saved in the sequential matrix D, together
When the sequential matrix D is divided by several two-dimentional sub-blocks according to preset target size, and will be drawn in the sequential matrix D
The each sub-block divided is converted into column vector all in accordance with the mode that column stack, and is to index from dictionary with the element in each column vector
Corresponding vector is picked out in the Φ of space constitutes dictionary subspace Φi, i=1,2 ..., k draw dictionary space to realize
Point；
Atom collection updates step: finding out redundancy vector rt-1In each dictionary subspace Φ of divisioniIn projectionAnd calculate PiEnergyThe corresponding dictionary subspace of ceiling capacity is found out simultaneously
Index subscriptAccording to calculated index subscript λtFind out corresponding dictionary subspaceAnd benefit
Atom collection A is updated with dictionary subspace, i.e.,Wherein, AtIt is the atom collection currently updated, At-1It is
The atom collection that last iteration updates, after updating atom collection A each time, by λtCorresponding dictionary subspace0 is set to ensure
Do not repeat value；
Sparse coefficient updates step: utilizing the updated atom collection A of current iterationtLeast-squares estimation is carried out to sparse coefficient, is asked
Signal is in atom collection A outtIn sparse coefficient componentAnd update redundancy vector rt=y-Axt；
Export step: according to update redundancy vector r obtainedtIt is iterated stop technology, as redundancy vector rtNorm be greater than
When preset constant ε, then jumps to the atom collection and update step progress next iteration, otherwise terminate iteration and output signal is dilute
The final estimated value of sparse coefficient.
2. the compressed sensing image recovery method assisted as described in claim 1 using objective attribute target attribute, which is characterized in that described
Subspace partition step specifically includes:
It is assumed that target size is B=a × b, then template size is set as a × b, generate the two of corresponding infrared small target image size
Tie up sequential matrix D ∈ Rw×h, wherein a, b are respectively target priori length and target priori width, and w, h are respectively infrared small target
The length and width of image, w divide exactly a, and h divides exactly b；
Element in two-dimentional sequential matrix D is the column stacking value of corresponding infrared small target image pixel point position, according to target ruler
Very little priori size divides two-dimentional sequential matrix D, obtains D={ B1,B2,...,Bi,...Bk, wherein
When the aliquant target priori length a of infrared small target image length w, the aliquant mesh of infrared small target picture traverse h
When marking priori width b, then according to the actual situation template size is set as that target can be completely covered, it also can be by infrared small mesh
The value that the length and width of logo image divide exactly；
The mode that each sub-block divided in the sequential matrix D stacks all in accordance with column is converted into column vector, with each
Element in column vector is that index picks out corresponding vector composition dictionary subspace Φ from the Φ of dictionary spacei, i=1,
2 ..., k, i.e. division dictionary subspace, Φi=Φ (Vec (Bi)), i=1,2 ..., k.
3. a kind of compressed sensing image recovery system assisted using objective attribute target attribute, which is characterized in that the system comprises:
Initialization module is used for redundancy vector rt-1, atom collection A and the number of iterations t these parameters are initialized, wherein
rt-1Decrement direction finding the amount y, A for being initialized as input are initialized as empty set, and t is initialized as 1；
Subspace partition module, for generating and infrared small target image sequential matrix D ∈ R of the same sizew×h, wherein w is
The length of infrared small target image, h are the width of infrared small target image, and 1 is saved in the sequential matrix D to w × h's
Number, while the sequential matrix D is divided by several two-dimentional sub-blocks according to preset target size, and by the sequential matrix D
Middle divided each sub-block is converted into column vector all in accordance with the mode that column stack, and is index with the element in each column vector
Corresponding vector is picked out from the Φ of dictionary space constitutes dictionary subspace Φi, i=1,2 ..., k, to realize to dictionary sky
Between division；
Atom collection update module, for finding out redundancy vector rt-1In each dictionary subspace Φ of divisioniIn projectionAnd calculate PiEnergyThe corresponding dictionary subspace of ceiling capacity is found out simultaneously
Index subscriptAccording to calculated index subscript λtFind out corresponding dictionary subspaceAnd it utilizes
Dictionary subspace is updated atom collection A, i.e.,Wherein, AtIt is the atom collection currently updated, At-1On being
The atom collection that an iteration updates, after updating atom collection A each time, by λtCorresponding dictionary subspace0 is set to ensure not
Repeat value；
Sparse coefficient update module, for utilizing the updated atom collection A of current iterationtLeast square is carried out to sparse coefficient to estimate
Meter, finds out signal in atom collection AtIn sparse coefficient componentAnd update redundancy vector rt=y-
Axt；
Output module, for according to update redundancy vector r obtainedtIt is iterated stop technology, as redundancy vector rtNorm
When greater than preset constant ε, then jumps to the atom collection and update step progress next iteration, otherwise terminate iteration and export letter
The final estimated value of number sparse coefficient.
4. the compressed sensing image recovery system assisted as claimed in claim 3 using objective attribute target attribute, which is characterized in that described
Subspace partition module is specifically used for:
It is assumed that target size is B=a × b, then template size is set as a × b, generate the two of corresponding infrared small target image size
Tie up sequential matrix D ∈ Rw×h, wherein a, b are respectively target priori length and target priori width, and w, h are respectively infrared small target
The length and width of image, w divide exactly a, and h divides exactly b；
Element in two-dimentional sequential matrix D is the column stacking value of corresponding infrared small target image pixel point position, according to target ruler
Very little priori size divides two-dimentional sequential matrix D, obtains D={ B1,B2,...,Bi,...Bk, wherein
When the aliquant target priori length a of infrared small target image length w, the aliquant mesh of infrared small target picture traverse h
When marking priori width b, then according to the actual situation template size is set as that target can be completely covered, it also can be by infrared small mesh
The value that the length and width of logo image divide exactly；
The mode that each sub-block divided in the sequential matrix D stacks all in accordance with column is converted into column vector, with each
Element in column vector is that index picks out corresponding vector composition dictionary subspace Φ from the Φ of dictionary spacei, i=1,
2 ..., k, i.e. division dictionary subspace, Φi=Φ (Vec (Bi)), i=1,2 ..., k.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610647440.5A CN106296611B (en) | 2016-08-09 | 2016-08-09 | A kind of compressed sensing image recovery method and its system using objective attribute target attribute auxiliary |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610647440.5A CN106296611B (en) | 2016-08-09 | 2016-08-09 | A kind of compressed sensing image recovery method and its system using objective attribute target attribute auxiliary |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106296611A CN106296611A (en) | 2017-01-04 |
| CN106296611B true CN106296611B (en) | 2019-04-16 |
Family
ID=57667399
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610647440.5A Expired - Fee Related CN106296611B (en) | 2016-08-09 | 2016-08-09 | A kind of compressed sensing image recovery method and its system using objective attribute target attribute auxiliary |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106296611B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018027584A1 (en) * | 2016-08-09 | 2018-02-15 | 深圳大学 | Method and system for restoring image using target attribute assisted compression perception |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103198500A (en) * | 2013-04-03 | 2013-07-10 | 西安电子科技大学 | Compressed sensing image reconstruction method based on principal component analysis (PCA) redundant dictionary and direction information |
| CN105811992A (en) * | 2016-03-01 | 2016-07-27 | 深圳大学 | Compressed sensing method and system capable of separating sparse signals |
-
2016
- 2016-08-09 CN CN201610647440.5A patent/CN106296611B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103198500A (en) * | 2013-04-03 | 2013-07-10 | 西安电子科技大学 | Compressed sensing image reconstruction method based on principal component analysis (PCA) redundant dictionary and direction information |
| CN105811992A (en) * | 2016-03-01 | 2016-07-27 | 深圳大学 | Compressed sensing method and system capable of separating sparse signals |
Non-Patent Citations (2)
| Title |
|---|
| Sub-sampled IFFT based compressive sampling;Liang Zhongyin等;《2015 IEEE Region 10 Conference》;20151104;正文第1-4页 |
| 基于秩极小化的压缩感知图像恢复算法;沈燕飞等;《电子学报》;20160331;第44卷(第3期);第572-578页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106296611A (en) | 2017-01-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10706593B2 (en) | Method and system for image reconstruction using target attribute assisted compressive sensing | |
| Hawe et al. | Dense disparity maps from sparse disparity measurements | |
| US20190303731A1 (en) | Target detection method and device, computing device and readable storage medium | |
| CN108510535A (en) | A kind of high quality depth estimation method based on depth prediction and enhancing sub-network | |
| CN106663316A (en) | Block sparse compressive sensing-based infrared image reconstruction method and system thereof | |
| Zhang et al. | 3d crowd counting via multi-view fusion with 3d gaussian kernels | |
| CN104077761B (en) | Multi-focus image fusion method based on self-adaption sparse representation | |
| Deng et al. | Noisy depth maps fusion for multiview stereo via matrix completion | |
| Cheng et al. | Supermatching: Feature matching using supersymmetric geometric constraints | |
| CN108832934A (en) | A kind of two-dimensional quadrature match tracing optimization algorithm based on singular value decomposition | |
| CN110501072A (en) | A kind of reconstructing method of the fast illuminated spectrum imaging system based on the constraint of tensor low-rank | |
| CN103674244A (en) | Rapid deconvolution two-dimensional fiber spectrum extraction method based on GPU | |
| CN107547088A (en) | Enhanced self-adapted segmentation orthogonal matching pursuit method based on compressed sensing | |
| Nakos et al. | Stronger L2/L2 compressed sensing; without iterating | |
| Han et al. | Boosting target-level infrared and visible image fusion with regional information coordination | |
| CN106296611B (en) | A kind of compressed sensing image recovery method and its system using objective attribute target attribute auxiliary | |
| Wang et al. | A modified algorithm based on smoothed l0 norm in compressive sensing signal reconstruction | |
| Wu et al. | Cross-fiber spatial-temporal co-enhanced networks for video action recognition | |
| Zhu et al. | Accurate, scalable and parallel structure from motion | |
| Zografos et al. | Fast and accurate motion segmentation using linear combination of views | |
| CN104616304A (en) | Self-adapting support weight stereo matching method based on field programmable gate array (FPGA) | |
| Sun et al. | Cycle-SfM: Joint self-supervised learning of depth and camera motion from monocular image sequences | |
| CN103606189B (en) | A kind of track base system of selection towards non-rigid three-dimensional reconstruction | |
| CN107886555A (en) | A kind of radio interference array distributed image method for reconstructing | |
| CN106652023B (en) | A kind of method and system of the extensive unordered quick exercise recovery structure of image |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190416 |