CN105574824A - Multi-target genetic optimization compressed sensing image reconstruction method based on ridgelet dictionary - Google Patents

Abstract

The invention discloses a multi-target genetic optimization compressed sensing image reconstruction method based on a ridgelet dictionary, and mainly aims to solve the problem of inaccurate image reconstruction due to inaccurate sparsity prediction in an existing reconstruction method. An implementation process of the method comprises the following steps: (1) discriminating structures of image blocks, and marking the image blocks as smooth image blocks, single-direction image blocks and multi-direction image blocks; (2) clustering observation vectors corresponding to different types of image blocks, and constructing initial groups for every type of image blocks; (3) performing crossover based on variable length codes at different gene positions on the current groups, and performing insertion and multipoint variation operation of new individuals based on prior information on the crossover groups; (4) calculating a non-dominated solution set of the current groups, and calculating an inflection point to serve as an optimal solution; and (5) calculating estimated values of the image blocks with the optimal solution, and sequentially splicing the image blocks into a whole image. Through adoption of the method, sparsity being relatively suitable for the image blocks can be found; an optimal atom combination under the sparsity is found; and high reconstruction quality and high robustness are achieved. The method can be applied to reconstruction of natural images and medical images.

Description

Based on the multi-objective Genetic optimization compressed sensing image reconstructing method of ridge ripple dictionary

Technical field

The invention belongs to technical field of image processing, relate to the non-convex compressed sensing image reconstructing method of multiple-objection optimization, can be used for the reconstruct to medical image and natural image.

Background technology

Compressive sensing theory is by the sparse characteristic of exploitation signal, and under the condition much smaller than Nyquist Nyquist sampling rate, stochastic sampling obtains the discrete sample of signal, then by non-linear reconstruction algorithm ideally reconstruction signal.The crucial part of compressive sensing theory is the combination realizing signal sampling and data compression, namely to the object also realizing while signal sampling compressing, with far below Nyquist sampling rate to signal sampling, so both save resource and bandwidth, alleviate again the pressure that higher bandwidth is brought to signal collecting device.Under this new signal processing model, the frequency of sampling no longer determine by the size of signal bandwidth, but relevant with the factor such as the uncorrelated degree of sparse transformation base, the sparse degree of signal with observing matrix.

The non-convex compressed sensing image reconstruction algorithm based on ridge ripple dictionary learning of current proposition, utilize evolutionary programming algorithm to learn the more excellent atom combination obtained on dictionary direction, this optimum atom combination is finally used to carry out reconstructed image, but this class algorithm all needs to know degree of rarefication in advance, namely reconstruct the number of the atom combination Atom of needs, but in actual applications, degree of rarefication is unknown often.

At present based in the non-convex compressed sensing image reconstruction algorithm of ridge ripple dictionary learning, usually fixing degree of rarefication, namely fix the atom number required for reconstruct signal, ask optimum atom combination, this is irrational, because the atom number required for a real signal reconstruction and degree of rarefication uncertain often, thus exist and estimate inaccurate due to degree of rarefication and cause the inaccurate problem of Image Reconstruction.

Summary of the invention

The object of the invention is to the deficiency for above-mentioned prior art, propose a kind of multi-objective Genetic optimization compressed sensing image reconstructing method based on ridge ripple dictionary, to improve the Image Reconstruction accuracy under degree of rarefication unknown condition.

For achieving the above object, technical scheme of the present invention is as follows:

(1) the Random Orthogonal Gauss observing matrix Φ of transmit leg transmission and the measurement vector y to each block that image block compression sampling obtains is received;

(2) structure discrimination is carried out to the image block corresponding to each observation vector y, image block is labeled as smooth piece, one direction block and multi-direction piece, and the direction of record direction block;

(3) to smooth piece, one direction block, multi-direction piece of corresponding measurement vector carries out cluster with neighbour's propagation clustering AP algorithm respectively;

(4) corresponding to each class observation vector image block, arranging Population Size is N, if current evolutionary generation is I, the coding by two targets of each individuality in population is set to forming, wherein first aim correspondence wants the degree of rarefication of reconstructed blocks, namely the number of atom combination Atom is reconstructed, second target correspondence wants the dictionary atom of reconstructed image block to combine, according to the classification wanting reconstructed image block to be labeled, initialization is carried out to population at individual, and calculate the fitness value of each individuality in population, obtain initial population;

(5) interlace operation based on the variable-length encoding of different genes position is performed to current population at individual, and upgrade the first aim of corresponding population at individual, obtain the population after intersecting;

(6) population after current intersection is performed based on the update of the new individuality of prior imformation, obtain new population and calculate the fitness value of new population individuality;

(7) multiple spot mutation operation is performed to the second target of current new population individuality;

(8) according to first aim and the corresponding fitness value of each individuality of new population, non-dominant disaggregation F is calculated;

(9) according to the number of new population individuality in non-dominant disaggregation F, uniform selection population at individual of future generation;

(10) judge whether current population iterations I reaches maximum iteration time or reach fitness requirement, if reach iterations or fitness requirement, according to the non-dominant disaggregation F of population, find the population at individual at flex point place, then perform step (11) as optimum individual; Otherwise iterations increases 1, i.e. I=I+1, returns step (5);

(11) according to the second target of optimum individual, calculate the estimated value of all image blocks, and be combined into a complete image by piecemeal order and export.

Compared with prior art, tool has the following advantages in the present invention:

1. the present invention carries out the initialization of population according to direction and yardstick, make population with some preferably solution start to search optimum solution, providing larger possibility for finding optimum solution, decreasing and finding time of optimum solution.

2. be encoded in population by degree of rarefication together with atom combination in the present invention, simultaneously by optimizing, the optimum solution that available multiobject method finds degree of rarefication and atom combination to trade off, improves the accuracy of reconstruct.

3. the present invention adopts variable-length encoding to population, carries out the interlace operation of random different point of crossing, makes to produce more multifarious solution, is conducive to searching the compromise optimum solution of degree of rarefication.

4. the present invention is newly individual at the position radom insertion that degree of rarefication is sparse, the while that this making to add population multifarious, is more conducive to finding optimum solution, decreases the time finding optimum solution.

Accompanying drawing explanation

Fig. 1 is realization flow figure of the present invention;

Fig. 2 uses the present invention under 30% sampling rate to the reconstruction result figure of Lena figure;

Fig. 3 uses the present invention under 30% sampling rate to the reconstruction result figure of Barbara figure.

Embodiment

With reference to Fig. 1, implementation step of the present invention is as follows:

Step 1, receives the Random Orthogonal Gauss observing matrix Φ of transmit leg transmission and the measurement vector y to each block that image block compression sampling obtains.

Step 2, carries out structure discrimination to the image block corresponding to each observation vector y, image block is labeled as smooth piece, one direction block and multi-direction piece, and the direction of record direction block.

(2a) according to following formula, corresponding noise vector is calculated to the measurement vector y of each image block

$\hat{x}={\mathrm{\Φ}}^{+}(y-\mathrm{\Φ}d{\left(\mathrm{\Φ}d\right)}^{+}y)$

In formula, y is the observation vector of image to be determined block, and Φ is the Gaussian matrix for observing, and d is a value is 1 column vector entirely, () ⁺it is the pseudo inverse matrix calculating matrix;

(2b) to each noise vector calculating noise value ξ, wherein vector two norms square;

(2c) judge whether ξ is less than threshold value 0.05, just the image block of this measurement correspondence is labeled as smooth piece if be less than; Otherwise just do not mark;

(2d) image block will do not made marks, then be labeled as one direction block and multi-direction piece in accordance with the following steps:

(2d1) existing ridge ripple is crossed complete dictionary and be divided into 36 sub-dictionary Ψ by direction ₁, Ψ ₂..., Ψ _i... Ψ ₃₆, to each observation vector y, be reconstructed on this little dictionary with OMP algorithm respectively, each sub-dictionary Ψ corresponding in the sequence of calculation _iobservation residual error r:

$r=\left|\right|y-{\mathrm{\ΦD}}_r\[{\left({\mathrm{\ΦD}}_r\right)}^{+}y\]|{|}_2^2$

So obtain an observation residual sequence r ₁, r ₂..., r _i... r ₃₆, find minimum value position i in the sequence, i=1 in sequence, 2 ..., 36, in formula, y is the observation vector of image to be determined block, D _rsub-dictionary Ψ _iin the combination of 10 atoms maximum with y correlativity, () ⁺it is the pseudo inverse matrix calculating matrix;

(2d2) utilize position i-2 in sequence, i-1, i, i+1, i+2 find five corresponding residual values r _i-2, r _i-1, r _i, r _i+1, r _i+2; If i is 1, then r _i-2and r _i-1use r respectively ₃₆and r ₃₅replace; If i is 2, then r _i-2use r ₃₆replace; If i is 36, then r _i+1and r _i+2use r respectively ₁and r ₂replace; If i is 35, then r _i+2use r ₁replace;

(2d3) corresponding to observation vector y image block marks: if r _i-2be greater than r _i-1, r _i-1be greater than 1.2r _i, r _i+1be greater than 1.2r _i, and r _i+2be greater than r _i+1, then image block corresponding for observation vector y is labeled as one direction, and i is appointed as the direction of this image block, otherwise just this image block is labeled as multi-direction.

Step 3, to smooth piece, one direction block, multi-direction piece of corresponding measurement vector carries out cluster with neighbour's propagation clustering AP algorithm respectively.

Clustering method has a variety of, such as C means clustering method, fuzzy clustering method, affine clustering method, and neighbour's propagation clustering AP algorithm etc., in the present embodiment, the clustering method of use is neighbour's propagation clustering AP algorithm.Neighbour's propagation clustering AP algorithm, AffinityPropagation (AP) cluster is a kind of new clustering algorithm proposed on Science magazine for 2007.It carries out cluster according to the similarity between N number of data point, does not need to specify clusters number in advance.

It is implemented as follows:

(3a) corresponding to all smooth image blocks observation vector carries out cluster, obtains the cluster of smooth image block;

(3b) corresponding to all one direction image blocks observation vector carries out cluster, obtains the cluster of one direction image block;

(3c) corresponding to all multidirectional image blocks observation vector carries out cluster, obtains the cluster of multidirectional image block.

Step 4, the image block corresponding to each class observation vector, carries out initialization according to the classification wanting reconstructed image block to be labeled to population at individual, and calculates the fitness value of each individuality in population, obtains initial population.

(4a) arrange Population Size and degree of rarefication scope: for each class smooth image block, arranging Population Size N is 20, and degree of rarefication scope is [3,50]; For each class one direction image block, arranging Population Size N is 20, and degree of rarefication scope is [10,60]; For each class multidirectional image block, arranging Population Size N is 36, and degree of rarefication scope is [10,60];

(4b) initialization population P={ (s ₁, g ₁), (s ₂, g ₂) ..., (s _k, g _k) ... (s _n, g _n), k=1,2 ..., N, wherein, s _kthe first aim of each individuality of corresponding population, g _kthe second target of each individuality of corresponding population;

(4b1) the N number of positive integer { s of random generation within the scope of degree of rarefication ₁, s ₂..., s _k..., s _n, with the first aim s of each individuality of this N number of positive integer initialization population _k;

(4b2) according to the first aim value s planting each individuality _k, the second target g of each individuality of initialization population _k:

For smooth piece, take out the atom of front 5 yardsticks in each direction in ridge ripple dictionary 36 directions, by direction composition director dictionary, take out s at random at every turn _kindividual director dictionary, at random at this s _ks is taken out in individual director dictionary _kindividual atom composed atom combination, the second target g of initialization population at individual _k;

For one direction block, from ridge ripple dictionary, take out the sub-dictionary Ψ of sub-dictionary as this one direction class image block in m direction and 4 directions adjacent with left and right, m direction _g, Ψ _g={ Ψ ^m-2, Ψ ^m-1, Ψ ^m, Ψ ^m+1, Ψ ^m+2, wherein, m represents the direction of this one direction class image block, at random at sub-dictionary Ψ _gmiddle taking-up s _kindividual atom composed atom combination, the second target g of initialization population at individual _k;

For multi-direction class image block, take out the atom in ridge ripple dictionary 36 directions by direction, composition direction dictionary Ψ ₁, Ψ ₂..., Ψ _k... Ψ ₃₆, to a kth population at individual, on a kth director dictionary, find s at random _kindividual atom composed atom combination, the second target g of an initialization kth population at individual _k;

(4d) to all individualities of initial population, by following fitness function, the fitness value of each individuality is calculated:

$f\left(g_k\right)=\underset{j=1}{\overset{n}{\Σ}}\left|\right|Y_j-{\mathrm{\ΦD}}_j{A}_j|{|}_2^2$

Wherein, f (g _k) be the fitness value on a kth population at individual of the image block that a certain class observation vector that will reconstruct is corresponding, n is the measurement number comprised in the observation vector of a certain class that will reconstruct, and j is the numbering measured in a certain class, j=1,2 ..., a certain class after n, Y cluster measures vector, Y _jthat in the observation vector of a certain class that will reconstruct, jth measures vector, D _jthe ridge ripple dictionary atom combination of a jth image block of a certain class image block of reconstruct, A _jthe rarefaction representation coefficient of the image block that in the observation vector of a certain class that will reconstruct, a jth observation vector is corresponding, be vectorial two norms square.

Step 5, performs the interlace operation based on the variable-length encoding of different genes position to current population at individual, and upgrades the first aim of corresponding population at individual, obtains the population after intersecting.

Interlace operation in the past, usually a random selected integer is as point of crossing, then the atom after point of crossing is exchanged, two that obtain two new individual the same with original individual lengths, perform the interlace operation of different point of crossing in this example, obtain two and the different new individuality of original individual lengths, it is implemented as follows:

(5a) go out T to population at individual T=N × C according to crossover probability C Stochastic choice, each selection a pair population at individual intersects;

(5b), when intersecting, stochastic generation is less than two positive integers of selected population at individual first aim, as two point of crossing; Again by the second target of two population at individuals, i.e. atom combination, the atom behind point of crossing exchanges, and forms two new population at individuals;

(5c) number of the individual second target Atom of two new populations is calculated, respectively as the first aim of two new population individualities;

(5d) first aim of new population individuality is greater than to the individuality of the degree of rarefication upper range limit of setting, using the first aim of the upper bound of degree of rarefication scope as it, again according to the direction of second target Atom, select atom combination as its second target;

(5d1) direction of individual second target Atom is first added up;

(5d2) from each direction, first an atom is respectively selected;

(5d3) judge whether the atom number selected reaches requirement, just stop, otherwise continue (5d2) step if reach requirement, the atom finally selected combination is as new individual second target.

Step 6, performs based on the update of the new individuality of prior imformation the population after current intersection, obtains new population and calculates the fitness value of new population individuality.

(6a) first arranging new degree of rarefication value set E is empty set, and the first aim taking out all population at individuals is saved in set G;

(6b) all elements in G is arranged from small to large, calculate the spacing value between two adjacent elements;

(6c) judge to have do not have the spacing value between two adjacent elements to be greater than 2, just therefrom two maximum elements of spacing value are selected if had, the intermediate value calculating these two elements joins in new degree of rarefication set E, and this value is also joined in original degree of rarefication value set G;

(6d) step (6b) is continued until do not have spacing value between the degree of rarefication of two vicinities to be greater than 2.

(6e) by the value in new degree of rarefication set E, as the new first aim inserting population at individual, and then from corresponding sub-dictionary, the second target of atom composed atom combination as them is selected at random according to the classification that image block is labeled;

If what the image block reconstructed was labeled is smooth piece, the atom then taking out front 5 yardsticks of all directions from whole ridge ripple dictionary forms a yardstick dictionary, the integer atom composed atom combination corresponding to individual first aim is selected at random, as the second target of individuality in yardstick dictionary;

If what the image block reconstructed was labeled is one direction block, then from corresponding director dictionary, select the integer atom composed atom combination corresponding to individual first aim at random, as the second target of individuality;

If what the image block reconstructed was labeled is multi-direction piece, then from whole ridge ripple dictionary, select the integer atom composed atom combination corresponding to individual first aim at random, as the second target of individuality;

(6d) fitness value of each new population individuality is calculated according to the second target of new population individuality.

Step 7, performs multiple spot mutation operation to the second target of current population at individual.

(7a) according to mutation probability C ' Stochastic choice B population at individual B=N × C ';

(7b) when making a variation, for the second target of each selected population at individual, namely in atom combination a position arbitrarily atom ridge ripple dictionary in the atom that goes out of random choose replace, produce a new population at individual;

(7c) fitness of the new population individuality that variation produces is calculated;

(7d) compare the fitness of parent and filial generation, if the fitness of filial generation is less than parent, just replaces parent with filial generation, otherwise do not replace.

Step 8, according to first aim and the corresponding fitness value of each individuality of new population, calculates non-dominant disaggregation F.

(8a) set non-domination solution F as empty set, domination disaggregation D is empty set;

(8b) to each population at individual p _t, t=1,2 ..., N, finds first aim to be less than its population at individual, is recorded in set H;

(8c) population at individual p is compared _twith the fitness value of population at individual in set H, if there is a population at individual p in set H _h, h=1,2 ..., N, h ≠ t, its fitness is less than p _tfitness, then population at individual p _tby population at individual p _harranged, by population at individual p _tjoin in domination disaggregation Q; Otherwise, p _tnot arrange by any individuality, it is joined in non-domination solution F;

(8d) judge that in population, each is individual according to this, finally obtain non-domination solution F and domination disaggregation Q.

Step 9, according to the number of population at individual in non-dominant disaggregation F, uniform selection population at individual of future generation.

If the number of population at individual just equals Population Size in non-dominant disaggregation F, then directly select population at individual in non-dominant disaggregation F as population of future generation;

If the number of population at individual is less than Population Size in non-dominant disaggregation F, then first select the population at individual in all non-dominant disaggregation F, then select remaining population at individual from domination disaggregation Q, obtain population of future generation;

If the number of population at individual is greater than Population Size in non-dominant disaggregation F, then in non-dominant disaggregation F according to following steps, selected population individuality enters population of future generation:

The first step: by population at individual p minimum for the first aim in non-dominant disaggregation F _awith the population at individual p that first aim is maximum _bjoin population of future generation in, and establishing v=2, v is population of future generation the sum of middle population at individual, w reduction 2, w is the population at individual sum in non-dominant disaggregation F;

Second step: in calculating non-dominant disaggregation F, remaining population at individual is to population the minimum euclidean distance b of middle individuality ₁, b ₂..., b _e..., b _w-v, e=1,2 ..., w-v;

3rd step: at Euclidean distance b ₁, b ₂..., b _e..., b _w-vpopulation at individual p corresponding to middle maximizing _c, joined population of future generation in, v increases by 1, i.e. v=v+1; W reduces 1, i.e. w=w-1;

4th step: judge to join population of future generation whether the sum of middle individuality reaches requirement, just stops, otherwise continue second step if reach requirement.

Step 10, according to the non-dominant disaggregation F of population, finds the population at individual at flex point place.

(10a) according to first aim and the second target of each population at individual in non-dominant disaggregation F, a two-dimensional coordinate system is set up: the first aim of population at individual is as horizontal ordinate; Fitness value corresponding to the second target of population at individual is as ordinate;

(10b) the population at individual p finding first aim minimum _awith the population at individual p that first aim is maximum _b;

(10c) calculate except population at individual p _aand p _ball population at individuals are in addition to p _aand p _bthe distance of line, then apart from maximum corresponding population at individual, is exactly the population at individual of flex point position.

Step 11, judges whether current population iterations I reaches maximum iteration time or reach fitness requirement, if reach iterations or fitness requirement, the second target retaining optimum individual performs step (12); Otherwise iterations increases 1, i.e. I=I+1, returns step (5).

Step 12, according to the second target of optimum individual, calculates the estimated value of all image blocks, and is combined into a complete image by piecemeal order and exports.

(12a) estimated value of all image blocks is calculated by following formulae discovery:

${\mathrm{\α}}_j=\[{\left(\mathrm{\Φ}{\tilde{D}}_j\right)}^{+}{Y}_j\]$

$x_j={\tilde{D}}_j{\mathrm{\α}}_j$

Wherein, j is the numbering measured in a certain class, j=1,2 ..., n, x _jrepresent the estimated value of a jth image block of a certain class image block, represent the optimum atom combination wanting reconstructed image block, Y _jrepresent the jth observation vector in a certain class observation vector, () ⁺represent the pseudo inverse matrix of compute matrix.

Effect of the present invention is further illustrated by the image of following emulation and data:

1. simulated conditions

(1) emulation experiment use that the Lena in the standard testing image storehouse of 512 × 512 schemes, Barbara figure, tile size is decided to be 16 × 16;

(2) observing matrix of this experiment is random Gaussian observing matrix, sampling rate 30%;

(3) the Ridgelet redundant dictionary scale that this experiment adopts is 12032, altogether 36 directions;

(4) this tests smooth class, each iterations 100 times during multi-direction class reconstruct, one direction class iteration 50 times;

(5) this experimental selection CPU is Interi5-3470, and dominant frequency is 3.2GHZ, inside saves as 4G, and operating system is 64 Win7, and emulation platform is Matlab2012a.

2. emulate content and result

Emulation 1, be under the condition of 30% in sampling rate, by the inventive method, Lena figure be reconstructed, simulation result as shown in Figure 2, wherein:

The experimental result of Fig. 2 illustrates, uses the reconstructed image that the inventive method obtains

The partial enlarged drawing that Fig. 2 (a) is Fig. 2 (a) for the former figure of Lena, Fig. 2 (b);

The restructuring graph of Fig. 2 (c) for obtaining with the present invention, the partial enlarged drawing that Fig. 2 (d) is Fig. 2 (c);

The experimental result explanation of Fig. 2, use the reconstructed image that the inventive method obtains, image is smoother, as can be seen from the contrast of each partial enlarged drawing, the smooth part that the present invention goes out lena arm reconstructs very level and smooth, edge is very clear again, illustrates that the present invention have found the suitable degree of rarefication of reconstructed image block, better with this degree of rarefication quality reconstruction.

Emulation 2, be under the condition of 30% in sampling rate, by the inventive method, Barbara figure be reconstructed, simulation result as shown in Figure 3, wherein:

The partial enlarged drawing that Fig. 3 (a) is Fig. 3 (a) for the former figure of Barbara, Fig. 3 (b);

The restructuring graph of Fig. 3 (c) for obtaining with the present invention, the partial enlarged drawing that Fig. 3 (d) is Fig. 3 (c);

The experimental result of Fig. 3 illustrates, use the reconstructed image that the inventive method obtains, reconstruct very level and smooth to Barbara smooth part, Edge texture is clear, illustrates that the present invention have found the suitable degree of rarefication of reconstructed image block, better with this degree of rarefication quality reconstruction.

In sum, present invention achieves by degree of rarefication and atom combination are optimized as two targets simultaneously, obtain the atom combination of the suitable degree of rarefication of reconstructed image and optimum, obtain the compressed sensing quality reconstruction good to natural image.

Claims (10)

1., based on a multi-objective Genetic optimization compressed sensing image reconstructing method for ridge ripple dictionary, it is characterized in that: comprise the steps:

(1) the Random Orthogonal Gauss observing matrix Φ of transmit leg transmission and the measurement vector y to each block that image block compression sampling obtains is received;

(2) structure discrimination is carried out to the image block corresponding to each observation vector y, image block is labeled as smooth piece, one direction block and multi-direction piece, and the direction of record direction block;

(3) to smooth piece, one direction block, multi-direction piece of corresponding measurement vector carries out cluster with neighbour's propagation clustering AP algorithm respectively;

(4) corresponding to each class observation vector image block, arranging Population Size is N, if current evolutionary generation is I, the coding by two targets of each individuality in population is set to forming, wherein first aim correspondence wants the degree of rarefication of reconstructed blocks, namely the number of atom combination Atom is reconstructed, second target correspondence wants the dictionary atom of reconstructed image block to combine, according to the classification wanting reconstructed image block to be labeled, initialization is carried out to population at individual, and calculate the fitness value of each individuality in population, obtain initial population;

(5) interlace operation based on the variable-length encoding of different genes position is performed to current population at individual, and upgrade the first aim of corresponding population at individual, obtain the population after intersecting;

(6) population after current intersection is performed based on the update of the new individuality of prior imformation, obtain new population and calculate the fitness value of new population individuality;

(7) multiple spot mutation operation is performed to the second target of current new population individuality;

(8) according to first aim and the corresponding fitness value of each individuality of new population, non-dominant disaggregation F is calculated;

(9) according to the number of new population individuality in non-dominant disaggregation F, uniform selection population at individual of future generation;

(10) judge whether current population iterations I reaches maximum iteration time or reach fitness requirement, if reach iterations or fitness requirement, according to the non-dominant disaggregation F of population, find the population at individual at flex point place, then perform step (11) as optimum individual; Otherwise iterations increases 1, i.e. I=I+1, returns step (5);

(11) according to the second target of optimum individual, calculate the estimated value of all image blocks, and be combined into a complete image by piecemeal order and export.

2. the evolution multiple goal compressed sensing image reconstructing method based on ridge ripple dictionary according to claim 1, wherein in step (2), structure discrimination is carried out to the image block corresponding to each observation vector, image block is labeled as smooth piece, one direction block and multi-direction piece, and the direction of record direction block, carry out in accordance with the following steps:

(2a) first according to following formula, corresponding noise vector is calculated to the measurement vector y of each image block

$\hat{x}={\mathrm{\Φ}}^{+}(y-\mathrm{\Φ}d{\left(\mathrm{\Φ}d\right)}^{+}y)$

In formula, y is the observation vector of image to be determined block, and Φ is the Gaussian matrix for observing, and d is a value is 1 column vector entirely, () ⁺it is the pseudo inverse matrix calculating matrix;

(2b) to each noise vector calculating noise value ξ, wherein vector two norms square;

(2c) judge whether ξ is less than threshold value 0.05, just the image block of this measurement correspondence is labeled as smooth piece if be less than; Otherwise just do not mark;

(2d) image block will do not made marks, then be labeled as one direction block and multi-direction piece in accordance with the following steps:

(2d1) existing ridge ripple is crossed complete dictionary and be divided into 36 sub-dictionary Ψ by direction ₁, Ψ ₂..., Ψ _i... Ψ ₃₆, to each observation vector y, be reconstructed on this little dictionary with OMP algorithm respectively, each sub-dictionary Ψ corresponding in the sequence of calculation _iobservation residual error r:

$r=\left|\right|y-{\mathrm{\ΦD}}_r\[{\left({\mathrm{\ΦD}}_r\right)}^{+}y\]|{|}_2^2,$

So obtain an observation residual sequence r ₁, r ₂..., r _i... r ₃₆, find minimum value position i in the sequence, i=1 in sequence, 2 ..., 36, in formula, y is the observation vector of image to be determined block, D _rsub-dictionary Ψ _iin the combination of 10 atoms maximum with y correlativity, () ⁺it is the pseudo inverse matrix calculating matrix;

(2d2) utilize position i-2 in sequence, i-1, i, i+1, i+2 find five corresponding residual values r _i-2, r _i-1, r _i, r _i+1, r _i+2; If i is 1, then r _i-2and r _i-1use r respectively ₃₆and r ₃₅replace; If i is 2, then r _i-2use r ₃₆replace; If i is 36, then r _i+1and r _i+2use r respectively ₁and r ₂replace; If i is 35, then r _i+2use r ₁replace;

(2d3) corresponding to observation vector y image block marks: if r _i-2be greater than r _i-1, r _i-1be greater than 1.2r _i, r _i+1be greater than 1.2r _i, and r _i+2be greater than r _i+1, then image block corresponding for observation vector y is labeled as one direction, and i is appointed as the direction of this image block, otherwise just this image block is labeled as multi-direction.

3. the evolution multiple goal compressed sensing image reconstructing method based on ridge ripple dictionary according to claim 1, wherein according to the classification wanting reconstructed image block to be labeled, initialization is carried out to population at individual in step (4), and calculate the fitness value of each individuality in population, carry out in accordance with the following steps:

(4a) arrange Population Size and degree of rarefication scope: for each class smooth image block, arranging Population Size N is 20, and degree of rarefication scope is [3,50]; For each class one direction image block, arranging Population Size N is 20, and degree of rarefication scope is [10,60]; For each class multidirectional image block, arranging Population Size N is 36, and degree of rarefication scope is [10,60];

(4b) initialization population P={ (s ₁, g ₁), (s ₂, g ₂) ..., (s _k, g _k) ... (s _n, g _n), k=1,2 ..., N, wherein, s _kthe first aim of each individuality of corresponding population, g _kthe second target of each individuality of corresponding population;

(4b1) the N number of positive integer { s of random generation within the scope of degree of rarefication ₁, s ₂..., s _k..., s _n, with the first aim s of each individuality of this N number of positive integer initialization population _k;

(4b2) according to the first aim value s planting each individuality _k, the second target g of each individuality of initialization population _k:

For smooth piece, take out the atom of front 5 yardsticks in each direction in ridge ripple dictionary 36 directions, by direction composition director dictionary, take out s at random at every turn _kindividual director dictionary, at random at this s _ks is taken out in individual director dictionary _kindividual atom composed atom combination, the second target g of initialization population at individual _k;

For one direction block, from ridge ripple dictionary, take out the sub-dictionary Ψ of sub-dictionary as this one direction class image block in m direction and 4 directions adjacent with left and right, m direction _g, Ψ _g={ Ψ ^m-2, Ψ ^m-1, Ψ ^m, Ψ ^m+1, Ψ ^m+2, wherein, m represents the direction of this one direction class image block, at random at sub-dictionary Ψ _gmiddle taking-up s _kindividual atom composed atom combination, the second target g of initialization population at individual _k;

For multi-direction class image block, take out the atom in ridge ripple dictionary 36 directions by direction, composition direction dictionary Ψ ₁, Ψ ₂..., Ψ _k... Ψ ₃₆, to a kth population at individual, on a kth director dictionary, find s at random _kindividual atom composed atom combination, the second target g of an initialization kth population at individual _k;

(4d) to all individualities of initial population, by following fitness function, the fitness value of each individuality is calculated:

$f\left(g_k\right)=\underset{j=1}{\overset{n}{\Σ}}\left|\right|Y_j-{\mathrm{\ΦD}}_j{A}_j|{|}_2^2$

Wherein, f (g _k) be the fitness value on a kth population at individual of the image block that a certain class observation vector that will reconstruct is corresponding, n is the measurement number comprised in the observation vector of a certain class that will reconstruct, and j is the numbering measured in a certain class, j=1,2 ..., a certain class after n, Y cluster measures vector, Y _jthat in the observation vector of a certain class that will reconstruct, jth measures vector, D _jthe ridge ripple dictionary atom combination of a jth image block of a certain class image block of reconstruct, A _jthe rarefaction representation coefficient of the image block that in the observation vector of a certain class that will reconstruct, a jth observation vector is corresponding, be vectorial two norms square.

4. the evolution multiple goal compressed sensing image reconstructing method based on ridge ripple dictionary according to claim 1, wherein in step (5), the interlace operation based on the variable-length encoding of different genes position is performed to current population at individual, and upgrade the first aim of corresponding population at individual, carry out in accordance with the following steps:

(5a) go out T to population at individual T=N × C according to crossover probability C Stochastic choice, each selection a pair population at individual intersects;

(5b), when intersecting, stochastic generation is less than two positive integers of selected population at individual first aim, as two point of crossing; Again by the second target of two population at individuals, i.e. atom combination, the atom behind point of crossing exchanges, and forms two new population at individuals;

(5c) number of the individual second target Atom of two new populations is calculated, respectively as the first aim of two new population individualities;

(5d) first aim of new population individuality is greater than to the individuality of the degree of rarefication upper range limit of setting, using the first aim of the upper bound of degree of rarefication scope as it, again according to the direction of second target Atom, select atom combination as its second target;

(5d1) direction of individual second target Atom is first added up;

(5d2) from each direction, first an atom is respectively selected;

(5d3) judge whether the atom number selected reaches requirement, just stop, otherwise continue (5d2) step if reach requirement, the atom finally selected combination is as new individual second target.

5. the evolution multiple goal compressed sensing image reconstructing method based on ridge ripple dictionary according to claim 1, wherein in step (6), the update based on the new individuality of prior imformation is performed to current population, calculate the fitness value of new population individuality again, carry out in accordance with the following steps:

(6a) first arranging new degree of rarefication value set E is empty set, and the first aim taking out all population at individuals is saved in set G;

(6b) all elements in G is arranged from small to large, calculate the spacing value between two adjacent elements;

(6c) judge to have do not have the spacing value between two adjacent elements to be greater than 2, just therefrom two maximum elements of spacing value are selected if had, the intermediate value calculating these two elements joins in new degree of rarefication set E, and this value is also joined in original degree of rarefication value set G;

(6d) step (6b) is continued until do not have spacing value between the degree of rarefication of two vicinities to be greater than 2.

(6e) by the value in new degree of rarefication set E, as the new first aim inserting population at individual, and then from corresponding sub-dictionary, the second target of atom composed atom combination as them is selected at random according to the classification that image block is labeled;

If what the image block reconstructed was labeled is smooth piece, the atom then taking out front 5 yardsticks of all directions from whole ridge ripple dictionary forms a yardstick dictionary, the integer atom composed atom combination corresponding to individual first aim is selected at random, as the second target of individuality in yardstick dictionary;

If what the image block reconstructed was labeled is one direction block, then from corresponding director dictionary, select the integer atom composed atom combination corresponding to individual first aim at random, as the second target of individuality;

If what the image block reconstructed was labeled is multi-direction piece, then from whole ridge ripple dictionary, select the integer atom composed atom combination corresponding to individual first aim at random, as the second target of individuality;

(6d) fitness value of each new population individuality is calculated according to the second target of new population individuality.

6. the evolution multiple goal compressed sensing image reconstructing method based on ridge ripple dictionary according to claim 1, wherein performs multiple spot mutation operation to the second target of current population at individual in step (7), carries out in accordance with the following steps:

(7a) according to mutation probability C ' Stochastic choice B population at individual B=N × C ';

(7b) when making a variation, for the second target of each selected population at individual, namely in atom combination a position arbitrarily atom ridge ripple dictionary in the atom that goes out of random choose replace, produce a new population at individual;

(7c) fitness of the new population individuality that variation produces is calculated;

(7d) compare the fitness of parent and filial generation, if the fitness of filial generation is less than parent, just replaces parent with filial generation, otherwise do not replace.

7. the evolution multiple goal compressed sensing image reconstructing method based on ridge ripple dictionary according to claim 1, wherein calculates non-domination solution collection F, carries out in accordance with the following steps in step (8):

(8a) set non-domination solution F as empty set, domination disaggregation Q is empty set;

(8b) to each population at individual p _t, t=1,2 ..., N, finds first aim to be less than its population at individual, is recorded in set H;

(8c) population at individual p is compared _twith the fitness value of population at individual in set H, if there is a population at individual p in set H _h, h=1,2 ..., N, h ≠ t, its fitness is less than p _tfitness, then population at individual p _tby population at individual p _harranged, by population at individual p _tjoin in domination disaggregation Q; Otherwise, p _tnot arrange by any individuality, it is joined in non-domination solution F;

(8c) judge that in population, each is individual according to this, finally obtain non-domination solution F and domination disaggregation Q.

8. the evolution multiple goal compressed sensing image reconstructing method based on ridge ripple dictionary according to claim 1, wherein in step (9) according to the number of population at individual in non-dominant disaggregation F, uniform selection new population of future generation is individual, is carry out according to the number and Population Size that compare population at individual in non-dominant disaggregation F:

If the number of population at individual just equals Population Size in non-dominant disaggregation F, then directly select population at individual in non-dominant disaggregation F as population of future generation;

If the number of population at individual is less than Population Size in non-dominant disaggregation F, then first select the population at individual in all non-dominant disaggregation F, then select remaining population at individual from domination disaggregation Q, obtain population of future generation;

If the number of population at individual is greater than Population Size in non-dominant disaggregation F, then in non-dominant disaggregation F according to following steps, selected population individuality enters population of future generation:

The first step: by population at individual p minimum for the first aim in non-dominant disaggregation F _awith the population at individual p that first aim is maximum _bjoin population of future generation in, and establishing v=2, v is population of future generation the sum of middle population at individual, w reduction 2, w is the population at individual sum in non-dominant disaggregation F;

Second step: in calculating non-dominant disaggregation F, remaining population at individual is to population the minimum euclidean distance b of middle individuality ₁, b ₂..., b _e..., b _w-v, e=1,2 ..., w-v;

3rd step: at Euclidean distance b ₁, b ₂..., b _e..., b _w-vpopulation at individual p corresponding to middle maximizing _c, joined population of future generation in, v increases by 1, i.e. v=v+1; W reduces 1, i.e. w=w-1;

4th step: judge to join population of future generation whether the sum of middle individuality reaches requirement, just stops, otherwise continue second step if reach requirement.

9. the evolution multiple goal compressed sensing image reconstructing method based on ridge ripple dictionary according to claim 1, wherein in step (10) according to the non-dominant disaggregation F of population, find the population at individual at flex point place, carry out in accordance with the following steps:

(10a) according to first aim and the second target of each population at individual in non-dominant disaggregation F, a two-dimensional coordinate system is set up: the first aim of population at individual is as horizontal ordinate; Fitness value corresponding to the second target of population at individual is as ordinate;

(10b) the population at individual p finding first aim minimum _awith the population at individual p that first aim is maximum _b;

(10c) calculate except population at individual p _aand p _ball population at individuals are in addition to p _aand p _bthe distance of line, then apart from maximum corresponding population at individual, is exactly the population at individual of flex point position.

10. the evolution multiple goal compressed sensing image reconstructing method based on ridge ripple dictionary according to claim 1, the second target according to optimum individual wherein described in step (11), calculate the estimated value of all image blocks, carry out according to following formula:

$A_j=\[{\left(\mathrm{\Φ}{\tilde{D}}_j\right)}^{+}{Y}_j\]$

$X_j={\tilde{D}}_j{A}_j$

Wherein, A _jbe the rarefaction representation coefficient of the image block that in the observation vector of a certain class that will reconstruct, a jth observation vector is corresponding, j is the numbering measured in a certain class, j=1,2 ..., n, X _jrepresent the estimated value of a jth image block of a certain class image block, represent the optimum atom combination wanting reconstructed image block, Y _jrepresent the jth observation vector in a certain class observation vector, () ⁺represent the pseudo inverse matrix of compute matrix.