CN107492129A - Non-convex compressed sensing optimal reconfiguration method with structuring cluster is represented based on sketch - Google Patents
Non-convex compressed sensing optimal reconfiguration method with structuring cluster is represented based on sketch Download PDFInfo
- Publication number
- CN107492129A CN107492129A CN201710707916.4A CN201710707916A CN107492129A CN 107492129 A CN107492129 A CN 107492129A CN 201710707916 A CN201710707916 A CN 201710707916A CN 107492129 A CN107492129 A CN 107492129A
- Authority
- CN
- China
- Prior art keywords
- image block
- mrow
- particle
- sketch
- block
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T9/00—Image coding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
- H03M7/00—Conversion of a code where information is represented by a given sequence or number of digits to a code where the same, similar or subset of information is represented by a different sequence or number of digits
- H03M7/30—Compression; Expansion; Suppression of unnecessary data, e.g. redundancy reduction
- H03M7/3059—Digital compression and data reduction techniques where the original information is represented by a subset or similar information, e.g. lossy compression
- H03M7/3062—Compressive sampling or sensing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Compression Of Band Width Or Redundancy In Fax (AREA)
Abstract
The invention discloses a kind of non-convex compressed sensing optimal reconfiguration method for representing to cluster with structuring based on sketch, mainly solves the problems, such as that the reconstruct of compression perceptual image is inaccurate under low sampling rate, its implementation process is：According to the sketch map of image, define can sketch block and can not sketch block, wherein can not sketch block include smooth piece and texture block, can sketch block include one direction and multi-direction piece；One direction block uses the cluster instructed based on sketch direction；Multi-direction piece of cluster of the use based on directional spreding feature；Smooth piece is clustered with texture block using gray scale；More measurement vector observations are carried out to every a kind of image block；During reconstruct, use and final reconstructed image is obtained based on the particle cluster algorithm intersected and atomic orientation constrains according to more calculation matrix of every a kind of image block, classification index and directional information, the present invention is compared with TS_RS and NR_DG methods, reconstructed image quality is high, robustness is good, the reconstruct available for natural image.
Description
Technical field
The invention belongs to technical field of image processing, and in particular to a kind of to represent non-convex with structuring cluster based on sketch
Compressed sensing optimal reconfiguration method, is reconstructed available for natural image.
Background technology
In recent years, there is a kind of new data theory compressed sensing CS in field of signal processing, the theory is adopted in data
Compression is realized while collection, breaching tradition, how Kui gathers the limitation of this special sampling thheorem, and leather is brought for data acquisition technology
The change of life property so that before the theory has wide application in fields such as compression imaging system, military cryptology, wireless sensings
Scape.Compressive sensing theory is mainly in terms of the observation of rarefaction representation, signal including signal and the reconstruct of signal etc. three.Wherein set
The effective observation of meter and reconstructing method are successfully promoted CS theories and applied to the important of real data model and acquisition system
Link.
This source problem of compressed sensing reconstruct is the non-convex optimization problem of zero norm constraint.In " Liu F, Lin L, Jiao
L,et al.Nonconvex compressed sensing by nature-inspired optimization
algorithms.[J].IEEE Transactions on Cybernetics,2015,45(5):1042-1053. " in a text
A kind of non-convex compressed sensing image reconstructing method is proposed, this method uses two benches reconstruction model, respectively using genetic optimization
Algorithm and clonal selection algorithm obtain optimal atom combination of the image block on dictionary direction and yardstick displacement.This method is only right
Image block has carried out smooth and Non-smooth surface differentiation, bad to direction, texture image block quality reconstruction.
Afterwards, in " Lin L, Fang L, Jiao L, et al.The Overcomplete Dictionary-Based
Directional Estimation Model and Nonconvex Reconstruction Methods[J].IEEE
Transactions on Cybernetics,2017,PP(99):In the texts of 1-12. " one, disclose one kind and be based on direction estimation mould
The non-convex compressed sensing image reconstructing method of type, this method also two stage reconstructing method, but before reconstruct, using direction
Estimate model, judge picture block structure type and walking direction according to observation data, and design according to different structure types
Different reconstruct dictionaries and evolution strategy, not only accelerate the speed of restructing algorithm but also add the precision of reconstructed image.This
Although kind of a method is judged the structure type of image block and direction, in the case where sample rate is relatively low and forbidden
Really, so as to causing reconstruct under low sampling rate to lack accuracy and robustness.
Meanwhile the slow deficiency of reconstructed velocity be present in above two method, both based on genetic Optimization Algorithm and gram
The two benches optimization of grand selection algorithm, speed is slower, is unfavorable for applying in real time.
The content of the invention
In view of the above-mentioned deficiencies in the prior art, the technical problem to be solved by the present invention is that propose that one kind is based on sketch
The non-convex compressed sensing optimal reconfiguration method with structuring cluster is represented, not only improves reconstructed velocity, and improve low sampling rate
Accuracy, the robustness of hypograph reconstruct.
The present invention uses following technical scheme：
Non-convex compressed sensing optimal reconfiguration method with structuring cluster is represented based on sketch, determined according to the sketch map of image
Justice can sketch block and can not sketch block, can not sketch block include smooth piece and texture block, can sketch block include one direction and in many ways
To block；One direction block uses the cluster instructed based on sketch direction；Multi-direction piece of cluster of the use based on directional spreding feature；Light
Sliding block and texture block are clustered using gray scale；More measurement vector observations are carried out to every a kind of image block；According to every a kind of image block
More calculation matrix, classification index and directional information are used and are reconstructed based on the particle cluster algorithm intersected and atomic orientation constrains
To final reconstructed image.
Preferably, comprise the following steps that：
Image block is divided into one direction image block, multi-direction figure by S1, data sender according to the sketch characteristic of original image
As four kinds of block, texture image block and smooth image block structure types；
S2, clustered according to the direction of one direction image block using based on the clustering method that direction is instructed, obtain folk prescription
To the cluster result of image block；
S3, using the clustering method based on directional spreding feature, the sketch block according to corresponding to multidirectional image block is gathered
Class, obtain multidirectional cluster result；
S4, using aggregation gray feature texture image block and smooth image block are clustered, obtain texture image block and
The cluster result of smooth image block；
S5, more measurement vector observations are carried out to every a kind of image block using random Gaussian calculation matrix Φ, obtain measure more
Vector observation set of matrices { Y1,Y2...,Yi,...,YC, YiFor the i-th class image block XiMore measurement vector observation matrixes, Yi=
ΦXi, i=i=1,2 ..., C, C be total classification number；
S6, vector observation set of matrices { Y will be measured more1,Y2...,Yi,...,YC, classification index vector l=(l1,
l2,...,ln,...,lN) and directional information vector z=(z1,z2,...,zi,...,zC) recipient is sent to, wherein, lnIt is n-th
Individual image block xnAffiliated class, ln∈ { 1,2 ..., C }, ziThe directional information of the i-th class image block is represented, if the i-th class image block
XiFor smooth piece, then zi=0, if the i-th class image block XiFor texture block or multi-direction piece, then zi=37, if the i-th class image
Block XiFor one direction class, and principal direction is equal to θ(l), then ziEqual to indexing l corresponding to principal direction；
S7, recipient judge the structure type per a kind of image block, and construct corresponding excessively complete according to the data received
Standby redundant dictionary；
S8, to every a kind of image block, respectively under excessively complete redundant dictionary corresponding to it, according to its corresponding more measurement vector
Observing matrix, optimal original of the particle on direction and yardstick is searched for using based on the particle cluster algorithm intersected and atomic orientation constrains
Sub-portfolio, and the estimate of image block is calculated；
S9, the estimate of all image blocks is spliced into a view picture weight again according to the classification index vector l information provided
Composition picture exports.
Preferably, step S1 is specially：
S11, the sketch map by initial sketch model acquisition original image；
S12, the sketch map of original image is divided into the big sketch block such as nonoverlapping, wherein there is the element that sketch line passes through
Retouch block be referred to as can sketch block, the image block that no sketch line passes through be referred to as can not sketch block；
S13, original picture block is divided into the big image block such as not overlapping, the size of image block and the size phase of sketch block
Together, with can the corresponding image block of sketch block be referred to as can sketch image block, with can not the corresponding image block of sketch block be referred to as can not element
Retouch image block；
S14, can not sketch image root tuber be divided into smooth image block and texture image block according to the size of each auto-variance, if not
Can sketch image block variance be less than threshold value T then the image block is smooth image block, otherwise the image block is texture image block；
S15, by can sketch image block according to the distribution situation of sketch line segment in corresponding sketch block be divided into one direction block and
Multi-direction piece, if can only have a sketch line segment in sketch block corresponding to sketch image block, or between sketch line segment
The deviation of directivity is no more than 15 °, then the image block is one direction block, and the direction of the one direction block is sketch in corresponding sketch block
The mean direction of line segment, otherwise the image block is multidirectional image block.
Preferably, step S2 is specially：
S21, amendment directional image block direction, make the directional information of image block and the excessively complete ridge ripple redundancy of structuring
The directional information of dictionary is matched, and the atom of excessively complete ridge ripple redundant dictionary is divided into 36 direction θl∈{θ(1),...,
θ(l),...,θ(36), θ(l)=(l-1) π/36, l=1,2 ..., 36, by the direction k of one direction image block to excessively complete ridge ripple
36 directions of redundant dictionary are drawn close, and obtain the principal direction of one direction image block
Wherein,| | the absolute value of expression,Represent | k-
θ(l)| θ when taking minimum value(l)Value；
S22, the one direction image block with identical principal direction is divided into one kind, one direction image block is according to respective different
Principal direction be divided into 36 groups, as 36 sub- direction classes；
S23, the image block in every sub- direction class is carried out second according to its gray feature clustered, obtain one direction figure
As the cluster result of block.
Preferably, step S3 is specially：
S31, sketch block corresponding to multi-direction piece is divided into the statistical window that nonoverlapping size is 4 × 4；
S32, the direction of each statistical window obtained by a direction pondization operation；
S33, the directional statistics distribution characteristics according to image block, calculate the otherness between two image blocks；
Itd is proposed in S34, the directional statistics distribution characteristics vector sum step S33 according to the image block proposed in step S32
Otherness computational methods between image block, are clustered to multidirectional image block, obtain multi-direction piece of first time cluster knot
Fruit；
S35, second of cluster is carried out according to its gray feature to every one kind after multidirectional image block for the first time cluster, obtained
To the final cluster result of multidirectional image block.
Preferably, in step S32, the concrete operations in direction pond are：
If S321, in statistical window only have a direction sketch line, then the direction of the window is sketch line pair
The direction answered；
If S322, the sketch line for including in a statistical window multiple directions, then in 4 × 4 statistical windows most
Direction of the direction of long sketch line segment as the window, pixel count of the sketch line segment in statistical window is as its length；
In step S33, the otherness between two image blocks is specially：
S331, the directional statistics distribution characteristics vector β for calculating a-th of image blockaWith the directional statistics point of b-th of image block
Cloth characteristic vector βbBetween angle difference vector D (βa, βb)：
Wherein,Represent theaIn individual image blockjThe direction of individual statistical window,Represent thebIn individual image blockjIndividual system
The direction of window is counted,Represent angle difference vector D (βa, βb)jIndividual component；
S332, according between two angle direction Statistical Distribution Characteristics vectors angle difference vector calculate two image blocks
Between otherness：
Preferably, step S7 is specially：
If S71, the i-th class image block directional information zi=0, then such is smooth image block class, and it is corresponding smooth
It is Ψ to cross complete redundant dictionarys, the excessively complete dictionary by ridge ripple redundant dictionary directive preceding 5 yardsticks ridge ripple redundancy
Sub- dictionary is formed, Represent comprising institute directive yardstick for h the sub- dictionary of ridge ripple redundancy, its
In, h=1,2 ..., 5；
If S72, the i-th class image block directional information zi∈ { 1,2 ..., 36 }, then such is smooth image block class, zi
Index l corresponding to such corresponding image block principal direction, take out the principal direction θ of the folk prescription block class(l)And the principal direction or so is adjacent
Excessively complete redundant dictionary of the sub- dictionary in 4 directions as the one direction block；
If S73, the i-th class image block directional information zi=37, then such is texture image block class or multi-direction figure
As block class, using the excessively complete redundant dictionary of whole ridge ripple as such excessively complete redundant dictionary.
Preferably, step S8 is specially：
S81, initialization population；
S82, the speed of atomic orientation constraint solve；
S83, particle position are according to particle update probability PrIt is updated, renewal operation adds for current particle position to be obtained
Particle rapidity after control；
S84, the new fitness value for calculating the particle after updating, if the new fitness value of particle is more than particle last time iteration
Fitness value, then replace current particle with the particle after renewal, and by the atom on each particle by atom numbering from it is small to
Big sequence, otherwise, particle keep constant, meanwhile, to the speed of more new particle by the index after the atomic order on particle again
Arrangement so that every one-dimensional and its speed of each particle position corresponds；
S85, by the use of the population currently moved as cross-species are treated, perform single-point crossover operation, set the probability to be
Pc, wherein, the crossover probability Pc=0.5 of one direction image block class and smooth piece of class, multidirectional image block class and texture image block
The crossover probability Pc=0.6 of class；The larger particle of a fitness is selected in the filial generation particle new from two, by its fitness value
Compared with producing the fitness value of its two parent particles：
If the fitness value of parent particle is big, pressed instead of parent particle, and by the atom on each particle after intersection
It is ranked up from small to large according to atom numbering, respective speed is rearranged according to the index after the atomic order on particle,
Otherwise, do not substitute；
The history optimal location and global history optimal location of particle in S86, more new particle；
S87, judge whether the iteration stopping condition that meets particle cluster algorithm, if satisfied, step S88 is performed, if not satisfied,
Return to step S82, continue to carry out particle velocity location renewal operation；
S88, the reconstruct base using the global history optimal solution of population as such image block；
S89, obtained all categories image block optimal atom combination, the estimation of all image blocks is calculated by following formula
Value xn：
xn=Dn[(ΦDn)+yn]
Wherein, DnRepresent the image block most has atom combination, ynRepresent the observation vector of the image block.
Preferably, step S81 is specially：
S811, to every a kind of one direction class image block, according to its excessively complete redundant dictionary Ψg5 different directions son
Dictionary, initialize a scale and be 20 population, and the population is divided into 5 groups, each group represents a direction, with
Machine initializes the particle in each group, and the atom on each particle numbered into ordered arrangement from small to large by atom, then by grain
The speed random initializtion of son is [- Lr,LrAny one value in] × 0.5, wherein, LrFor the excessively complete redundancy of the one direction class
The size of the sub- dictionary of correspondence direction r in dictionary, r ∈ { 1,2 ..., 36 }；
S812, to texture image block class and multidirectional image block class, initialize the population that a scale is 36, each grain
Filial generation table both direction, wherein, a direction is selected from the excessively complete redundant dictionary of ridge ripple and the particle numbering identical direction,
Another direction then randomly chooses one again from remaining direction, and the atom on each particle is also numbered from small certainly according to former
It is [- L to big ordered arrangement, then by the speed random initializtion of particlem,LmAny one value in] × 0.5, wherein, LmRepresent
Selected two represent the size of the excessively complete ridge ripple redundant dictionary in direction, and m represents that particle is numbered, m=1, and 2 ..., 36；
S813, according to the following formula, calculate smooth image block class, one direction image block class, texture image block class and multi-direction figure
As the fitness value of each particle in block class population：
Wherein, f (bm) be the population of image block corresponding to such more calculation matrix in m-th of particle fitness
Value, YiRepresent the i-th class image block XiMore measurement vector observation matrixes, bmIt is the solution in population representated by m-th of particle, it is right
Answer one group of base sub-portfolio in the excessively complete redundant dictionary of ridge ripple, ()+The pseudo inverse matrix of calculating matrix is represented,Represent
Square of matrix not this black norm of Luo Beini；
S814, the history optimal solution and global history optimal solution for initializing population：By each particle in population
History optimal solution is initialized as particle itself, and the maximum particle of fitness value, goes through as the overall situation in whole population of breaking forth
History optimal solution.
Preferably, step S82 is specially：
S821, the parameter in particle cluster algorithm is configured, is by the particle rapidity location updating of population：
Wherein,Speed corresponding to u-th of component of m-th of particle after the renewal of the t+1 times iteration is represented,Represent to work as
U-th of component of m-th of particle corresponding speed in preceding population,Represent u-th point of m-th of particle in current particle group
Amount,U-th of component of the history optimal location of m-th of particle in current particle group,Represent that current global history is optimal
Position gtU-th of component, u is positive integer, and t represents it is currently which time iteration, and w represents inertia weight, c1Represent Particle tracking
The weight coefficient of itself history optimal value, c2The weight coefficient of Particle tracking global history optimal value is represented, ξ, η are [0,1] areas
Interior equally distributed two random numbers；
S822, component is often tieed up according to current each particle firstOn atom where direction r, obtain r-th of direction
Atoms range [gr,hr]；Then the speed travel direction of particle is constrained according to below equation, the particle speed after being controlled
Degree
Compared with prior art, the present invention at least has the advantages that：
The present invention is represented based on sketch and the non-convex compressed sensing optimal reconfiguration method of structuring cluster, according to based on sketch
The structuring clustering method of expression, image block accurate structure type, classification and directional information are obtained, by these information together with every
The multivariable observed result of a kind of image block is together sent to recipient so that recipient can improve the accurate of Image Reconstruction
Property, use and constrained based on intersection and atomic orientation according to more calculation matrix of every a kind of image block, classification index and directional information
Particle cluster algorithm be reconstructed to obtain final reconstructed image, the present invention is compared with TS_RS and NR_DG methods, reconstructed image matter
Amount is high, and robustness is good, the reconstruct available for natural image.
Further, sketch characteristic of the present invention according to original image before observation is classified image block, is fully dug
The architectural feature of image block has been dug, the architectural feature of image block is being extracted according to observed result relative to prior art, is being obtained
Image block architectural feature it is more accurate, by the structure type of image block, classification and directional information and per a kind of multivariable
Observed result is together sent to recipient, and it is effective that these accurate prior informations can accurately instruct recipient to carry out image
Reconstruct, so as to lift the quality of Image Reconstruction and robustness.
Further, in natural image block comprising edge image block and smooth image block gray value mean square deviation all compared with
Low, it is difficult to be demarcated the two by a threshold value, the present invention utilizes the sketch characteristic of image, and being first divided into image block can element
Retouch block and can not sketch block, wherein can not sketch block include texture block and smooth piece, can sketch block include one direction block and in many ways
To block (i.e. the image block of bound edge edge), effectively the image block for including edge in natural image and smooth piece can be differentiated.
Further, a kind of clustering method based on directional spreding feature is proposed so that of a sort multidirectional image block
With similar direction structure feature, according to its corresponding sketch block, for different types of structure image block devise it is different
Clustering method, the clustering method based on gray scale is used for smooth piece and texture block, uses for one direction block and is referred to based on direction
The clustering method led, use the clustering method based on directional spreding feature for multidirectional image block so that the figure in same class
As block has similar direction structure feature.
Further, particle is searched on direction and yardstick with based on the particle cluster algorithm intersected and atomic orientation constrains
Optimal atom combination, and the estimate of image block is calculated, proposed in Optimization Solution a kind of based on intersection and courtyard direction
The particle cluster algorithm of constraint, improve the speed of Optimization Solution.
Further, the present invention uses particle cluster algorithm, improves the reconstructed velocity of image.
Below by drawings and examples, technical scheme is described in further detail.
Brief description of the drawings
Fig. 1 is the general flow chart of the present invention；
Fig. 2 is the schematic diagram that multi-direction piece of image block obtains directional statistics distribution；
Fig. 3 for the present invention and two kinds of existing methods under 20% sample rate respectively to the reconstruction result figure of Barbara figures, its
In, (a) is Barbara artworks, and (b) is figure a partial enlarged drawing, and (c) is the reconstruct image obtained with two benches reconstructing method,
(d) it is the partial enlarged drawing of figure (c), (e) is the reconstruct image that the reconstruct instructed with direction obtains, and (f) is figure e partial enlargement
Figure, (g) are the reconstruct image that the present invention obtains, and (h) is figure g partial enlarged drawing；
Fig. 4 for the present invention and two kinds of existing methods under 20% sample rate respectively to the reconstruction result figure of Lena figures, wherein,
(a) it is Lena artworks, (b) is figure a partial enlarged drawing, and (c) is the reconstruct image obtained with two benches reconstructing method, and (d) is figure c
Partial enlarged drawing, (e) is the obtained reconstruct image of reconstruct instructed with direction, and (f) is the partial enlarged drawing for scheming e, and (g) is this hair
Bright obtained reconstruct image, (h) are figure g partial enlarged drawing；
Fig. 5 for the present invention and two kinds of existing methods under different sample rates to the reconstitution time line chart of Lena figures.
Embodiment
The invention provides a kind of non-convex compressed sensing optimal reconfiguration method for representing to cluster with structuring based on sketch, root
According to the sketch map of image, define can sketch block and can not sketch block, wherein can not sketch block include smooth piece and texture block, can
Sketch block includes one direction and multi-direction piece；One direction block uses the cluster instructed based on sketch direction；Multi-direction piece uses base
In the cluster of directional spreding feature；Smooth piece is clustered with texture block using gray scale；More measurement vectors are carried out to every a kind of image block
Observation；During reconstruct, used according to more calculation matrix of every a kind of image block, classification index and directional information based on intersection and atom
The particle cluster algorithm of direction constraint obtains final reconstructed image.
Referring to Fig. 1, the present invention is represented based on sketch and the non-convex compressed sensing optimal reconfiguration method of structuring cluster, tool
Body step is as follows：
Image block is divided into one direction image block, multi-direction figure by S1, data sender according to the sketch characteristic of original image
As four kinds of block, texture image block and smooth image block different types of structure, and the direction of one direction image block is in structure type
It can be obtained during division.
S11, the sketch map by initial sketch model acquisition original image；
S12, the sketch map of original image is divided into the big sketch block such as nonoverlapping, wherein there is the element that sketch line passes through
Retouch block be referred to as can sketch block, the image block that no sketch line passes through be referred to as can not sketch block；
S13, original picture block is divided into the big image block such as not overlapping, the size of image block and the size phase of sketch block
Together, with can the corresponding image block of sketch block be referred to as can sketch image block, with can not the corresponding image block of sketch block be referred to as can not element
Retouch image block；
S14, can not sketch image root tuber be divided into smooth image block and texture image block according to the size of each auto-variance, if not
Can sketch image block variance be less than threshold value T then the image block is smooth image block, otherwise the image block is texture image block；
S15, by can sketch image block according to the distribution situation of sketch line segment in corresponding sketch block be divided into one direction block and
Multi-direction piece, if can only have a sketch line segment in sketch block corresponding to sketch image block, or between sketch line segment
The deviation of directivity is no more than 15 °, then the image block is one direction block, and the direction of the one direction block is sketch in corresponding sketch block
The mean direction of line segment, otherwise the image block is multidirectional image block.
S2, one direction image block use the clustering method instructed based on direction to be clustered according to its direction, obtain folk prescription
To the cluster result of image block；
Certain modification is made in S21, the direction to one direction image block so that the directional information of image block and the mistake of structuring
The directional information of complete ridge ripple redundant dictionary matches, and the atom of excessively complete ridge ripple redundant dictionary can be divided into 36 directionsθ(l)=(l-1) π/36, l=1,2 ..., 36, by the direction k of one direction image block to excessively complete
36 directions of standby ridge ripple redundant dictionary are drawn close, and obtain the principal direction of one direction image block
Wherein,| | the absolute value of expression,Represent | k-
θ(l)| θ when taking minimum value(l)Value；
S22, the one direction image block with identical principal direction is divided into one kind, such one direction image block is according to each
Different principal direction is divided into 36 groups, and we are called 36 sub- direction classes；
S23, the image block in every sub- direction class is carried out second according to its gray feature clustered, obtain one direction figure
As the cluster result of block.
S3, multidirectional image root tuber are gathered according to its corresponding sketch block using the clustering method based on directional spreding feature
Class, multidirectional cluster result is obtained, as shown in Fig. 2 detailed process is；
S31, sketch block corresponding to multi-direction piece is divided into the statistical window that nonoverlapping size is 4 × 4；
S32, the direction of each statistical window obtained by a direction pondization operation, the concrete operations in direction pond are：
If S321, in statistical window only have a direction sketch line, then the direction of the window is sketch line pair
The direction answered；
If S322, the sketch line for including in a statistical window multiple directions, then in 4 × 4 statistical windows most
Direction of the direction of long sketch line segment as the window, pixel count of the sketch line segment in statistical window is as its length；
In the sketch block all statistical windows direction composition correspondence image block directional statistics distribution characteristics vector β=
[β1,...,βj...,βJ], wherein βjIt is the direction of j-th of statistical window in correspondence image block, J is to be counted in each image block
The number of window, if in j-th of window without sketch line not as corresponding βj=0；
S33, the directional statistics distribution characteristics according to image block, calculate the otherness between two image blocks：
S331, the directional statistics distribution characteristics vector β for calculating a-th of image blockaWithbThe directional statistics of individual image block point
Cloth characteristic vector βbBetween angle difference vector D (βa, βb)：
Wherein,Represent theaIn individual image blockjThe direction of individual statistical window,Represent thebIn individual image blockjIndividual system
The direction of window is counted,Represent angle difference vector D (βa, βb)jIndividual component；
S332, according between two angle direction Statistical Distribution Characteristics vectors angle difference vector calculate two image blocks
Between otherness
Wherein, the otherness diff (β between two image blocksa, βb) smaller, similitude is higher between two image blocks；
Itd is proposed in S34, the directional statistics distribution characteristics vector sum step S33 according to the image block proposed in step S32
Otherness computational methods between image block, are clustered to multidirectional image block, obtain multi-direction piece of first time cluster knot
Fruit；
S35, second of cluster is carried out according to its gray feature to every one kind after multidirectional image block for the first time cluster, obtained
To the final cluster result of multidirectional image block.
S4, texture image block and smooth image block using the cluster of aggregation gray feature, obtain texture image block and light
The cluster result of sliding image block；
S5, more measurement vector observations are carried out to every a kind of image block using random Gaussian calculation matrix Φ, obtain measure more
Vector observation set of matrices { Y1,Y2...,Yi,...,YC, YiFor the i-th class image block XiMore measurement vector observation matrixes, Yi=
ΦXi, i=i=1,2 ..., C, C be total classification number；
S6, vector observation set of matrices { Y will be measured more1,Y2...,Yi,...,YC, classification index vector l=(l1,
l2,...,ln,...,lN), wherein lnIt is n-th image block xnAffiliated class, ln∈ { 1,2 ..., C }, and directional information vector z
=(z1,z2,...,zi,...,zC) it is sent to recipient, wherein ziThe directional information of the i-th class image block is represented, if the i-th class figure
As block XiFor smooth piece, then zi=0, if the i-th class image block XiFor texture block or multi-direction piece, then zi=37, if the i-th class
Image block XiFor one direction class, and principal direction is equal to θ(l), then ziEqual to indexing l corresponding to principal direction.
S7, recipient judge the structure type per a kind of image block, and construct corresponding excessively complete according to the data received
Standby redundant dictionary：
If S71, the i-th class image block directional information zi=0, then such is smooth image block class, and it is corresponding smooth
It is Ψ to cross complete redundant dictionarys, the excessively complete dictionary by ridge ripple redundant dictionary directive preceding 5 yardsticks ridge ripple redundancy
Sub- dictionary is formed, Represent comprising institute directive yardstick for h the sub- dictionary of ridge ripple redundancy, its
In, h=1,2 ..., 5；
If S72, the i-th class image block directional information zi∈ { 1,2 ..., 36 }, then such is smooth image block class, zi
Index l corresponding to such corresponding image block principal direction, take out the principal direction θ of the folk prescription block class(l)And the principal direction or so is adjacent
Excessively complete redundant dictionary of the sub- dictionary in 4 directions as the one direction block；
If S73, the i-th class image block directional information zi=37, then such is texture image block class or multi-direction figure
As block class, using the excessively complete redundant dictionary of whole ridge ripple as such excessively complete redundant dictionary.
S8, to every a kind of image block, respectively under excessively complete redundant dictionary corresponding to it, according to its corresponding more measurement vector
Observing matrix, optimal original of the particle on direction and yardstick is searched for using based on the particle cluster algorithm intersected and atomic orientation constrains
Sub-portfolio, and the estimate of image block is calculated.
S81, initialization population
S811, to every a kind of one direction class image block, according to its excessively complete redundant dictionary Ψg5 different directions son
Dictionary, initialize a scale and be 20 population, and the population is divided into 5 groups, each group represents a direction, with
Machine initializes the particle in each group, and the atom on each particle numbered into ordered arrangement from small to large by atom, then by grain
The speed random initializtion of son is [- Lr,LrAny one value in] × 0.5, wherein, LrFor the excessively complete redundancy of the one direction class
The size of the sub- dictionary of correspondence direction r in dictionary, r ∈ { 1,2 ..., 36 }；
S812, to texture image block class and multidirectional image block class, initialize the population that a scale is 36, each grain
Filial generation table both direction, one of direction is selected from the excessively complete redundant dictionary of ridge ripple and the particle numbering identical direction,
Another direction then randomly chooses one again from remaining direction, and the atom on each particle is also numbered from small certainly according to former
It is [- L to big ordered arrangement, then by the speed random initializtion of particlem,LmAny one value in] × 0.5, wherein LmRepresent
Selected two represent the size of the excessively complete ridge ripple redundant dictionary in direction, and m represents that particle is numbered, m=1, and 2 ..., 36；
S813, according to equation below, calculate smooth image block class, one direction image block class, texture image block class and multi-party
The fitness value of each particle into image block class population：
Wherein, f (bm) be the population of image block corresponding to such more calculation matrix in m-th of particle fitness
Value, bmIt is the solution in population representated by m-th of particle, correspond to one group of base subgroup in the excessively complete redundant dictionary of ridge ripple
Close, ()+The pseudo inverse matrix of calculating matrix is represented,Square of representing matrix not this black norm of Luo Beini；
S814, the history optimal solution and global history optimal solution for initializing population：By each particle in population
History optimal solution is initialized as particle itself, and the maximum particle of fitness value, goes through as the overall situation in whole population of breaking forth
History optimal solution；
S82, the speed of atomic orientation constraint solve
S821, the parameter in particle cluster algorithm is configured, is by the particle rapidity location updating of population：
Wherein,Speed corresponding to u-th of component of m-th of particle after the renewal of the t+1 times iteration is represented,Represent to work as
U-th of component of m-th of particle corresponding speed in preceding population,Represent u-th point of m-th of particle in current particle group
Amount,U-th of component of the history optimal location of m-th of particle in current particle group,Represent that current global history is optimal
Position gtU-th of component, u is positive integer, and maximum is equal to the degree of rarefication of setting, and t represents be currently which time iteration, represent
W inertia weights, w is from 0.9 linear decrease to 0.4, c1Represent the weight coefficient of Particle tracking itself history optimal value, c1From 2.5 lines
Property is decremented to 0.5, c2Represent the weight coefficient of Particle tracking global history optimal value, c2From 0.5 linear increment to 2.5, ξ, η are
Equally distributed two random numbers in [0,1] section；
Set up a particle position update probability parameter Pr, PrAs evolutionary generation is from 1.0 linear decreases to 0.1；
S822, component is often tieed up according to current each particle firstOn atom where direction r, obtain r-th of direction
Atoms range [gr,hr]；Then the speed travel direction of particle is constrained according to below equation, the particle speed after being controlled
Degree
S83, particle position are according to particle update probability PrIt is updated, wherein renewal operation adds for current particle position
Particle rapidity after being controlled；
S84, the new fitness value for calculating the particle after updating, if the new fitness value of particle is more than particle last time iteration
Fitness value, then replace current particle with the particle after renewal, and by the atom on each particle by atom numbering from it is small to
Big sequence, otherwise, particle keep constant, meanwhile, to the speed of more new particle by the index after the atomic order on particle again
Arrangement so that every one-dimensional and its speed of each particle position corresponds；
S85, by the use of the population currently moved as cross-species are treated, perform single-point crossover operation, set the probability to be
Pc, wherein the crossover probability Pc=0.5 of one direction image block class and smooth piece of class, multidirectional image block class and texture image block class
Crossover probability Pc=0.6；The larger particle of a fitness is selected in the filial generation particle new from two, by its fitness value with
The fitness value for producing its two parent particles is compared：If the fitness value of parent particle is big, instead of parent particle,
And be ranked up the atom on each particle after intersection from small to large according to atom numbering, by respective speed according to particle
On atomic order after index rearrange, otherwise, do not substitute；
The history optimal location and global history optimal location of particle in S86, more new particle；
S87, judge whether the iteration stopping condition that meets particle cluster algorithm, if satisfied, step S88 is performed, if not satisfied,
Return to step S82, continue to carry out particle velocity location renewal operation；
S88, the reconstruct base using the global history optimal solution of population as such image block.
S89, obtained all categories image block optimal atom combination, all image blocks are calculated by below equation
Estimate：
xn=Dn[(ΦDn)+yn]
Wherein, DnRepresent the y for most having atom combination, representing the image block of the image blocknObservation vector.
S9, the estimate of all image blocks is spliced into a view picture weight again according to the classification index vector l information provided
Composition picture exports.
Embodiment
1st, simulated conditions：The emulation of the present invention is in windows 7, SPI, CPU Intel (R) Core (TM) i5-3470, base
This frequency 3.20GHz, software platform are to be run on Matlab R2011b, and emulate selection is 512 × 512 four width standards survey
Try natural image Lena, Barbara, Boat, piecemeal size.
2nd, emulation content and result：
Emulation 1：
Under conditions of sample rate is 20%, weight is carried out to Barbara images respectively with the inventive method and existing method
Structure, for simulation result figure as shown in figure 3, wherein, Fig. 3 (a) is Barbara artworks, Fig. 3 (b) is Fig. 3 (a) partial enlarged drawing, is schemed
3 (c) is the reconstruct image obtained with two benches reconstructing method (TS_RS), and Fig. 3 (d) is Fig. 3 (c) partial enlarged drawing, and Fig. 3 (e) is
The reconstruct image that the reconstruct (NR_DG) instructed with direction obtains, Fig. 3 (f) are Fig. 3 (e) partial enlarged drawing, and Fig. 3 (g) is the present invention
Obtained reconstruct image, Fig. 3 (h) are Fig. 3 (g) partial enlarged drawing.
Reconstruct image Fig. 3 (g) and TS_RS of the present invention reconstruct image Fig. 3 (c), NR_DG reconstruct image Fig. 3 (e) are contrasted, with
Artwork Fig. 3 (a) is more like, partial enlarged drawing corresponding to contrast, and Fig. 3 (h) is than Fig. 3 (d), Fig. 3 (h), the line on Barbara trouser legs
Reason reconstructs apparent.
Fig. 3 experimental result explanation, the reconstructed image ratio obtained using the inventive method using two benches reconstructing method and
The reconstructed image that the reconstructing method that direction is instructed obtains, in visual effect more preferably.From pair of the partial enlarged drawing of these images
Reconstruct more clearly to the texture on Barbara trouser legs than can be seen that the present invention, illustrate of the invention edge to image,
One direction texture image block, which has, more accurately to be reconstructed.
Emulation 2：
Under conditions of sample rate is 20%, Lena images are reconstructed respectively with the inventive method and existing method,
Simulation result figure is as shown in figure 4, wherein, Fig. 4 (a) is Lena artworks, and Fig. 4 (b) is Fig. 4 (a) partial enlarged drawing, and Fig. 4 (c) is
The reconstruct image obtained with two benches reconstructing method (TS_RS), Fig. 4 (d) are Fig. 4 (c) partial enlarged drawing, and Fig. 4 (e) is to use direction
The reconstruct image that the reconstruct (NR_DG) of guidance obtains, Fig. 4 (f) are Fig. 4 (e) partial enlarged drawing, and Fig. 4 (g) present invention obtains
Reconstruct image, Fig. 4 (h) are Fig. 4 (g) partial enlarged drawing.
Reconstruct image Fig. 4 (g) and TS_RS of the present invention reconstruct image Fig. 4 (c), NR_DG reconstruct image Fig. 4 (e) are contrasted, with
Artwork Fig. 4 (a) is more like, partial enlarged drawing corresponding to contrast, and Fig. 4 (h) is than Fig. 4 (d), Fig. 4 (h), Lena shoulder parts edge
Become apparent from, smooth part has more preferable uniformity.
As seen in Figure 4, the accent image ratio obtained using the inventive method uses two benches reconstructing method and side
The reconstructed image obtained to the reconstructing method of guidance, become apparent from Lena shoulder parts edge, smooth part has more preferable
Uniformity, illustrate that there is preferable reconstruction property to natural image.
Emulation 3
Under different sample rates, progress is schemed to Lena, Barbar and Boat respectively with the inventive method and existing method
Reconstruct, and obtained numerical result is just compared, as a result as shown in table 1.
As it can be seen from table 1 PSNR the and SSIM values of reconstructed image of the present invention are than two benches reconstructing method (TS_RS) and side
It is high to the reconstructing method (NR_DG) of guidance, illustrate that the present invention has preferable reconstruction property to natural image.
Table 1 is the result of image Y-PSNR PSNR (structural similarity SSIM) index of three kinds of methods
Fig. 5 be the present invention and two kinds of existing methods under different sample rates to the reconstitution time line chart of Lena figures, can be with
Find out, the triangle broken line for representing the present invention be located under other two methods, illustrates under each sample rate of the invention need
Reconstitution time is most short, and the further instruction present invention to there is faster reconstructed velocity naturally.
To sum up, clearly reconstructed image can be obtained well under sampling proposed by the present invention and reconstructing method, with showing
There is observation to be compared with reconstructing method, the present invention improves the reconstruction quality and reconstructed velocity of image.
The technological thought of above content only to illustrate the invention, it is impossible to protection scope of the present invention is limited with this, it is every to press
According to technological thought proposed by the present invention, any change done on the basis of technical scheme, claims of the present invention is each fallen within
Protection domain within.
Claims (10)
1. the non-convex compressed sensing optimal reconfiguration method with structuring cluster is represented based on sketch, it is characterised in that according to image
Sketch map definition can sketch block and can not sketch block, can not sketch block include smooth piece and texture block, can sketch block include singly
Direction and multi-direction piece；One direction block uses the cluster instructed based on sketch direction；It is special that multi-direction piece of use is based on directional spreding
The cluster of sign；Smooth piece is clustered with texture block using gray scale；More measurement vector observations are carried out to every a kind of image block；According to each
More calculation matrix, classification index and the directional information of class image block are used based on the particle cluster algorithm intersected and atomic orientation constrains
It is reconstructed to obtain final reconstructed image.
A kind of 2. non-convex compressed sensing optimal reconfiguration side represented based on sketch with structuring cluster according to claim 1
Method, it is characterised in that comprise the following steps that：
S1, data sender according to the sketch characteristic of original image by image block be divided into one direction image block, multidirectional image block,
Four kinds of structure types of texture image block and smooth image block；
S2, clustered according to the direction of one direction image block using based on the clustering method that direction is instructed, obtain one direction figure
As the cluster result of block；
S3, using the clustering method based on directional spreding feature, the sketch block according to corresponding to multidirectional image block is clustered, and is obtained
To multidirectional cluster result；
S4, using aggregation gray feature texture image block and smooth image block are clustered, obtain texture image block and smooth
The cluster result of image block；
S5, more measurement vector observations are carried out to every a kind of image block using random Gaussian calculation matrix Φ, obtain more measurement vectors
Observing matrix set { Y1,Y2...,Yi,...,YC, YiFor the i-th class image block XiMore measurement vector observation matrixes, Yi=Φ Xi,
I=i=1,2 ..., C, C be total classification number；
S6, vector observation set of matrices { Y will be measured more1,Y2...,Yi,...,YC, classification index vector l=(l1,l2,...,
ln,...,lN) and directional information vector z=(z1,z2,...,zi,...,zC) recipient is sent to, wherein, lnIt is n-th image
Block xnAffiliated class, ln∈ { 1,2 ..., C }, ziThe directional information of the i-th class image block is represented, if the i-th class image block XiFor light
Sliding block, then zi=0, if the i-th class image block XiFor texture block or multi-direction piece, then zi=37, if the i-th class image block XiFor
One direction class, and principal direction is equal to θ(l), then ziEqual to indexing l corresponding to principal direction；
S7, recipient judge the structure type per a kind of image block, and construct corresponding excessively complete superfluous according to the data received
Remaining dictionary；
S8, to every a kind of image block, respectively under excessively complete redundant dictionary corresponding to it, according to its corresponding more measurement vector observation
Matrix, optimal atom group of the particle on direction and yardstick is searched for using based on the particle cluster algorithm intersected and atomic orientation constrains
Close, and the estimate of image block is calculated；
S9, the estimate of all image blocks is spliced into a view picture reconstruct image again according to the classification index vector l information provided
As output.
A kind of 3. non-convex compressed sensing optimal reconfiguration side represented based on sketch with structuring cluster according to claim 2
Method, it is characterised in that step S1 is specially：
S11, the sketch map by initial sketch model acquisition original image；
S12, the sketch map of original image is divided into the big sketch block such as nonoverlapping, wherein the sketch block for thering is sketch line to pass through
Referred to as can sketch block, the image block that no sketch line passes through be referred to as can not sketch block；
S13, original picture block being divided into the big image block such as not overlapping, the size of image block is identical with the size of sketch block,
With can the corresponding image block of sketch block be referred to as can sketch image block, with can not the corresponding image block of sketch block be referred to as can not sketch map
As block；
S14, can not sketch image root tuber be divided into smooth image block and texture image block according to the size of each auto-variance, if can not element
Retouch image block variance be less than threshold value T then the image block is smooth image block, otherwise the image block is texture image block；
S15, by can sketch image block be divided into one direction block and multi-party according to the distribution situation of sketch line segment in corresponding sketch block
To block, if can only have a sketch line segment, or the direction between sketch line segment in sketch block corresponding to sketch image block
Deviation is no more than 15 °, then the image block is one direction block, and the direction of the one direction block is sketch line segment in corresponding sketch block
Mean direction, otherwise the image block is multidirectional image block.
A kind of 4. non-convex compressed sensing optimal reconfiguration side represented based on sketch with structuring cluster according to claim 2
Method, it is characterised in that step S2 is specially：
S21, amendment directional image block direction, make the directional information of image block and the excessively complete ridge ripple redundant dictionary of structuring
Directional information match, the atom of excessively complete ridge ripple redundant dictionary is divided into 36 direction θl∈{θ(1),...,
θ(l),...,θ(36), θ(l)=(l-1) π/36, l=1,2 ..., 36, by the direction k of one direction image block to excessively complete ridge ripple
36 directions of redundant dictionary are drawn close, and obtain the principal direction of one direction image block
<mrow>
<mover>
<mi>k</mi>
<mo>&OverBar;</mo>
</mover>
<mo>=</mo>
<mi>arg</mi>
<munder>
<mrow>
<mi>m</mi>
<mi>i</mi>
<mi>n</mi>
</mrow>
<msup>
<mi>&theta;</mi>
<mrow>
<mo>(</mo>
<mi>l</mi>
<mo>)</mo>
</mrow>
</msup>
</munder>
<mrow>
<mo>|</mo>
<mrow>
<mi>k</mi>
<mo>-</mo>
<msup>
<mi>&theta;</mi>
<mrow>
<mo>(</mo>
<mi>l</mi>
<mo>)</mo>
</mrow>
</msup>
</mrow>
<mo>|</mo>
</mrow>
</mrow>
Wherein,| | the absolute value of expression,Represent | k- θ(l)| take
θ during minimum value(l)Value；
S22, the one direction image block with identical principal direction is divided into one kind, one direction image block is according to each different masters
Direction is divided into 36 groups, as 36 sub- direction classes；
S23, the image block in every sub- direction class is carried out second according to its gray feature clustered, obtain one direction image block
Cluster result.
A kind of 5. non-convex compressed sensing optimal reconfiguration side represented based on sketch with structuring cluster according to claim 2
Method, it is characterised in that step S3 is specially：
S31, sketch block corresponding to multi-direction piece is divided into the statistical window that nonoverlapping size is 4 × 4；
S32, the direction of each statistical window obtained by a direction pondization operation；
S33, the directional statistics distribution characteristics according to image block, calculate the otherness between two image blocks；
The image proposed in S34, the directional statistics distribution characteristics vector sum step S33 according to the image block proposed in step S32
Otherness computational methods between block, are clustered to multidirectional image block, obtain multi-direction piece of first time cluster result；
S35, second of cluster is carried out according to its gray feature to every one kind after multidirectional image block for the first time cluster, obtained more
The final cluster result of directional image block.
A kind of 6. non-convex compressed sensing optimal reconfiguration side represented based on sketch with structuring cluster according to claim 5
Method, it is characterised in that in step S32, the concrete operations in direction pond are：
If S321, in statistical window only have a direction sketch line, then the direction of the window be sketch line corresponding to
Direction；
If S322, the sketch line for including in a statistical window multiple directions, then most long in 4 × 4 statistical windows
Direction of the direction of sketch line segment as the window, pixel count of the sketch line segment in statistical window is as its length；
In step S33, the otherness between two image blocks is specially：
S331, the directional statistics distribution characteristics vector β for calculating a-th of image blockaIt is distributed with the directional statistics of b-th of image block special
Levy vectorial βbBetween angle difference vector D (βa, βb)：
<mrow>
<mi>D</mi>
<mrow>
<mo>(</mo>
<msup>
<mi>&beta;</mi>
<mi>a</mi>
</msup>
<mo>,</mo>
<msup>
<mi>&beta;</mi>
<mi>b</mi>
</msup>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mo>&lsqb;</mo>
<msubsup>
<mi>d</mi>
<mn>1</mn>
<mrow>
<mi>a</mi>
<mo>,</mo>
<mi>b</mi>
</mrow>
</msubsup>
<mo>,</mo>
<mo>...</mo>
<mo>,</mo>
<msubsup>
<mi>d</mi>
<mi>j</mi>
<mrow>
<mi>a</mi>
<mo>,</mo>
<mi>b</mi>
</mrow>
</msubsup>
<mo>,</mo>
<mo>...</mo>
<mo>,</mo>
<msubsup>
<mi>d</mi>
<mi>J</mi>
<mrow>
<mi>a</mi>
<mo>,</mo>
<mi>b</mi>
</mrow>
</msubsup>
<mo>&rsqb;</mo>
</mrow>
Wherein,The direction of j-th of statistical window in a-th of image block is represented,Represent j-th of statistics in b-th of image block
The direction of window,Represent angle difference vector D (βa, βb) j-th of component；
S332, according between two angle direction Statistical Distribution Characteristics vectors angle difference vector calculate two image blocks between
Otherness：
<mrow>
<mi>d</mi>
<mi>i</mi>
<mi>f</mi>
<mi>f</mi>
<mrow>
<mo>(</mo>
<msup>
<mi>&beta;</mi>
<mi>a</mi>
</msup>
<mo>,</mo>
<msup>
<mi>&beta;</mi>
<mi>b</mi>
</msup>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msup>
<mrow>
<mo>(</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>J</mi>
</munderover>
<msup>
<mrow>
<mo>(</mo>
<mrow>
<mi>s</mi>
<mi>i</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>d</mi>
<mi>j</mi>
<mrow>
<mi>a</mi>
<mo>,</mo>
<mi>b</mi>
</mrow>
</msubsup>
<mo>)</mo>
</mrow>
</mrow>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<mo>)</mo>
</mrow>
<mrow>
<mn>1</mn>
<mo>/</mo>
<mn>2</mn>
</mrow>
</msup>
<mo>.</mo>
</mrow>
A kind of 7. non-convex compressed sensing optimal reconfiguration side represented based on sketch with structuring cluster according to claim 2
Method, it is characterised in that step S7 is specially：
If S71, the i-th class image block directional information zi=0, then such is smooth image block class, and it is corresponding smooth excessively complete
Redundant dictionary is Ψs, the excessively complete dictionary by ridge ripple redundant dictionary directive preceding 5 yardsticks the sub- dictionary of ridge ripple redundancy
Form, Represent comprising institute directive yardstick for h the sub- dictionary of ridge ripple redundancy, wherein, h=
1,2,...,5；
If S72, the i-th class image block directional information zi∈ { 1,2 ..., 36 }, then such is smooth image block class, ziIt is corresponding
Index l corresponding to such image block principal direction, take out the principal direction θ of the folk prescription block class(l)And 4 that the principal direction or so is adjacent
Excessively complete redundant dictionary of the sub- dictionary in direction as the one direction block；
If S73, the i-th class image block directional information zi=37, then such is texture image block class or multidirectional image block
Class, using the excessively complete redundant dictionary of whole ridge ripple as such excessively complete redundant dictionary.
A kind of 8. non-convex compressed sensing optimal reconfiguration side represented based on sketch with structuring cluster according to claim 2
Method, it is characterised in that step S8 is specially：
S81, initialization population；
S82, the speed of atomic orientation constraint solve；
S83, particle position are according to particle update probability PrIt is updated, renewal operation adds for current particle position to be controlled
Particle rapidity afterwards；
S84, the new fitness value for calculating the particle after updating, if the new fitness value of particle is more than the suitable of particle last time iteration
Angle value is answered, then replaces current particle with the particle after renewal, and the atom on each particle is arranged from small to large by atom numbering
Sequence, otherwise, particle keep constant, meanwhile, the speed of more new particle is rearranged by the index after the atomic order on particle,
So that every one-dimensional and its speed of each particle position corresponds；
S85, by the use of the population currently moved as cross-species are treated, perform single-point crossover operation, setting probability is Pc, its
In, the crossover probability Pc=0.5 of one direction image block class and smooth piece of class, the friendship of multidirectional image block class and texture image block class
Pitch probability P c=0.6；The larger particle of a fitness is selected in the filial generation particle new from two, by its fitness value with producing
The fitness value of its two parent particles is compared：
If the fitness value of parent particle is big, instead of parent particle, and by the atom on each particle after intersection according to original
Son numbering is ranked up from small to large, and respective speed is rearranged according to the index after the atomic order on particle, otherwise,
Do not substitute；
The history optimal location and global history optimal location of particle in S86, more new particle；
S87, judge whether the iteration stopping condition that meets particle cluster algorithm, if satisfied, step S88 is performed, if not satisfied, returning
Step S82, continue to carry out particle velocity location renewal operation；
S88, the reconstruct base using the global history optimal solution of population as such image block；
S89, obtained all categories image block the combination of optimal atom, the estimate x of all image blocks is calculated by following formulan：
xn=Dn[(ΦDn)+yn]
Wherein, DnRepresent the image block most has atom combination, ynRepresent the observation vector of the image block.
A kind of 9. non-convex compressed sensing optimal reconfiguration side represented based on sketch with structuring cluster according to claim 8
Method, it is characterised in that step S81 is specially：
S811, to every a kind of one direction class image block, according to its excessively complete redundant dictionary Ψg5 different directions sub- dictionary,
One scale of initialization is 20 population, and the population is divided into 5 groups, and each group represents a direction, random first
Particle in each group of beginningization, and the atom on each particle is numbered into ordered arrangement from small to large by atom, then by particle
Speed random initializtion is [- Lr,LrAny one value in] × 0.5, wherein, LrFor the excessively complete redundant dictionary of the one direction class
In correspondence direction r sub- dictionary size, r ∈ { 1,2 ..., 36 }；
S812, to texture image block class and multidirectional image block class, initialize the population that a scale is 36, each particle generation
Table both direction, wherein, a direction is selected from the excessively complete redundant dictionary of ridge ripple and the particle numbering identical direction, another
Individual direction then randomly chooses one again from remaining direction, and the atom on each particle is also numbered from small to large certainly according to former
Ordered arrangement, then by the speed random initializtion of particle be [- Lm,LmAny one value in] × 0.5, wherein, LmSelected by expression
Two selected represent the size of the excessively complete ridge ripple redundant dictionary in direction, and m represents that particle is numbered, m=1, and 2 ..., 36；
S813, according to the following formula, calculate smooth image block class, one direction image block class, texture image block class and multidirectional image block
The fitness value of each particle in class population：
<mrow>
<mi>f</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>b</mi>
<mi>m</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mrow>
<mo>|</mo>
<mo>|</mo>
<msub>
<mi>Y</mi>
<mi>i</mi>
</msub>
<mo>-</mo>
<msub>
<mi>&Phi;b</mi>
<mi>m</mi>
</msub>
<mo>&lsqb;</mo>
<msup>
<mrow>
<mo>(</mo>
<msub>
<mi>&Phi;b</mi>
<mi>m</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>+</mo>
</msup>
<msub>
<mi>Y</mi>
<mi>i</mi>
</msub>
<mo>&rsqb;</mo>
<mo>|</mo>
<msubsup>
<mo>|</mo>
<mi>F</mi>
<mn>2</mn>
</msubsup>
</mrow>
</mfrac>
</mrow>
Wherein, f (bm) be the population of image block corresponding to such more calculation matrix in m-th of particle fitness value, YiTable
Show the i-th class image block XiMore measurement vector observation matrixes, bmIt is the solution in population representated by m-th of particle, correspond to ridge ripple
The one group of base sub-portfolio crossed in complete redundant dictionary, ()+The pseudo inverse matrix of calculating matrix is represented,Representing matrix is not
Square of this norm of Luo Beini crows；
S814, the history optimal solution and global history optimal solution for initializing population：By the history of each particle in population
Optimal solution is initialized as particle itself, the maximum particle of fitness value in whole population of breaking forth, as global history most
Excellent solution.
A kind of 10. non-convex compressed sensing optimal reconfiguration represented based on sketch with structuring cluster according to claim 8
Method, it is characterised in that step S82 is specially：
S821, the parameter in particle cluster algorithm is configured, is by the particle rapidity location updating of population：
<mrow>
<msubsup>
<mi>v</mi>
<mrow>
<mi>m</mi>
<mo>,</mo>
<mi>u</mi>
</mrow>
<mrow>
<mi>t</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
</msubsup>
<mo>=</mo>
<msubsup>
<mi>wv</mi>
<mrow>
<mi>m</mi>
<mo>,</mo>
<mi>u</mi>
</mrow>
<mi>t</mi>
</msubsup>
<mo>+</mo>
<msub>
<mi>c</mi>
<mn>1</mn>
</msub>
<mi>&xi;</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>p</mi>
<mrow>
<mi>m</mi>
<mo>,</mo>
<mi>u</mi>
</mrow>
<mi>t</mi>
</msubsup>
<mo>-</mo>
<msubsup>
<mi>b</mi>
<mrow>
<mi>m</mi>
<mo>,</mo>
<mi>u</mi>
</mrow>
<mi>t</mi>
</msubsup>
<mo>)</mo>
</mrow>
<mo>+</mo>
<msub>
<mi>c</mi>
<mn>2</mn>
</msub>
<mi>&eta;</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>g</mi>
<mi>u</mi>
<mi>t</mi>
</msubsup>
<mo>-</mo>
<msubsup>
<mi>b</mi>
<mrow>
<mi>m</mi>
<mo>,</mo>
<mi>u</mi>
</mrow>
<mi>t</mi>
</msubsup>
<mo>)</mo>
</mrow>
</mrow>
Wherein,Speed corresponding to u-th of component of m-th of particle after the renewal of the t+1 times iteration is represented,Represent current grain
U-th of component of m-th of particle corresponding speed in subgroup,U-th of component of m-th of particle in current particle group is represented,U-th of component of the history optimal location of m-th of particle in current particle group,Represent current global history optimal location
gtU-th of component, u is positive integer, and t represents it is currently which time iteration, and w represents inertia weight, c1Represent Particle tracking itself
The weight coefficient of history optimal value, c2The weight coefficient of Particle tracking global history optimal value is represented, ξ, η are in [0,1] section
Equally distributed two random numbers；
S822, component is often tieed up according to current each particle firstOn atom where direction r, obtain the original in r-th of direction
Subrange [gr,hr]；Then the speed travel direction of particle is constrained according to below equation, the particle rapidity after being controlled
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710707916.4A CN107492129B (en) | 2017-08-17 | 2017-08-17 | Non-convex compressive sensing optimization reconstruction method based on sketch representation and structured clustering |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710707916.4A CN107492129B (en) | 2017-08-17 | 2017-08-17 | Non-convex compressive sensing optimization reconstruction method based on sketch representation and structured clustering |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107492129A true CN107492129A (en) | 2017-12-19 |
| CN107492129B CN107492129B (en) | 2021-01-19 |
Family
ID=60646356
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710707916.4A Active CN107492129B (en) | 2017-08-17 | 2017-08-17 | Non-convex compressive sensing optimization reconstruction method based on sketch representation and structured clustering |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107492129B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109087367A (en) * | 2018-07-27 | 2018-12-25 | 西安航空学院 | A kind of high spectrum image Fast Compression sensing reconstructing method based on particle group optimizing |
| CN109451314A (en) * | 2018-04-23 | 2019-03-08 | 杭州电子科技大学 | A kind of compression of images cognitive method based on graph model |
| CN110570480A (en) * | 2019-07-19 | 2019-12-13 | 广东智媒云图科技股份有限公司 | Sketch drawing method of drawing robot, electronic equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105354800A (en) * | 2015-10-08 | 2016-02-24 | 西安电子科技大学 | Image structure-based particle swarm optimization non-convex compressed sensing image reconstruction method |
| CN105574824A (en) * | 2015-12-15 | 2016-05-11 | 西安电子科技大学 | Multi-target genetic optimization compressed sensing image reconstruction method based on ridgelet dictionary |
| CN106611420A (en) * | 2016-12-30 | 2017-05-03 | 西安电子科技大学 | SAR image segmentation method based on deconvolution network and sketch direction constraint |
| CN106651884A (en) * | 2016-12-30 | 2017-05-10 | 西安电子科技大学 | Sketch structure-based mean field variational Bayes synthetic aperture radar (SAR) image segmentation method |
-
2017
- 2017-08-17 CN CN201710707916.4A patent/CN107492129B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105354800A (en) * | 2015-10-08 | 2016-02-24 | 西安电子科技大学 | Image structure-based particle swarm optimization non-convex compressed sensing image reconstruction method |
| CN105574824A (en) * | 2015-12-15 | 2016-05-11 | 西安电子科技大学 | Multi-target genetic optimization compressed sensing image reconstruction method based on ridgelet dictionary |
| CN106611420A (en) * | 2016-12-30 | 2017-05-03 | 西安电子科技大学 | SAR image segmentation method based on deconvolution network and sketch direction constraint |
| CN106651884A (en) * | 2016-12-30 | 2017-05-10 | 西安电子科技大学 | Sketch structure-based mean field variational Bayes synthetic aperture radar (SAR) image segmentation method |
Non-Patent Citations (1)
| Title |
|---|
| 王蓉芳 等: "《利用纹理信息的图像分块自适应压缩感知》", 《电子学报》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109451314A (en) * | 2018-04-23 | 2019-03-08 | 杭州电子科技大学 | A kind of compression of images cognitive method based on graph model |
| CN109451314B (en) * | 2018-04-23 | 2021-06-08 | 杭州电子科技大学 | Image compression sensing method based on graph model |
| CN109087367A (en) * | 2018-07-27 | 2018-12-25 | 西安航空学院 | A kind of high spectrum image Fast Compression sensing reconstructing method based on particle group optimizing |
| CN109087367B (en) * | 2018-07-27 | 2022-09-27 | 西安航空学院 | High-spectrum image rapid compressed sensing reconstruction method based on particle swarm optimization |
| CN110570480A (en) * | 2019-07-19 | 2019-12-13 | 广东智媒云图科技股份有限公司 | Sketch drawing method of drawing robot, electronic equipment and storage medium |
| CN110570480B (en) * | 2019-07-19 | 2023-03-24 | 广东智媒云图科技股份有限公司 | Sketch drawing method of drawing robot, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107492129B (en) | 2021-01-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109960759B (en) | Recommendation system click rate prediction method based on deep neural network | |
| CN110188725A (en) | The scene Recognition system and model generating method of high-resolution remote sensing image | |
| Furukawa | SOM of SOMs | |
| CN112257597B (en) | Semantic segmentation method for point cloud data | |
| CN100557626C (en) | Image partition method based on immune spectrum clustering | |
| CN107220277A (en) | Image retrieval algorithm based on cartographical sketching | |
| CN107229904A (en) | A kind of object detection and recognition method based on deep learning | |
| CN107358293A (en) | A kind of neural network training method and device | |
| CN103824272B (en) | The face super-resolution reconstruction method heavily identified based on k nearest neighbor | |
| CN106874688A (en) | Intelligent lead compound based on convolutional neural networks finds method | |
| CN109658419A (en) | The dividing method of organella in a kind of medical image | |
| CN106228183A (en) | A kind of semi-supervised learning sorting technique and device | |
| CN106951911A (en) | A kind of quick multi-tag picture retrieval system and implementation method | |
| CN105608690A (en) | Graph theory and semi supervised learning combination-based image segmentation method | |
| CN112686097A (en) | Human body image key point posture estimation method | |
| CN104966105A (en) | Robust machine error retrieving method and system | |
| CN108257154A (en) | Polarimetric SAR Image change detecting method based on area information and CNN | |
| CN107492129A (en) | Non-convex compressed sensing optimal reconfiguration method with structuring cluster is represented based on sketch | |
| CN104751175B (en) | SAR image multiclass mark scene classification method based on Incremental support vector machine | |
| CN109711401A (en) | A kind of Method for text detection in natural scene image based on Faster Rcnn | |
| CN107122411A (en) | A kind of collaborative filtering recommending method based on discrete multi views Hash | |
| CN107832412A (en) | A kind of publication clustering method based on reference citation relation | |
| CN107247753A (en) | A kind of similar users choosing method and device | |
| CN105608713B (en) | A kind of bi-level image coding based on quaternary tree and efficient logical operation method | |
| Zhang et al. | Incomplete multiview nonnegative representation learning with multiple graphs |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |