CN109448065A - A kind of compression sensing method based on the measurement of gradient block adaptive - Google Patents
A kind of compression sensing method based on the measurement of gradient block adaptive Download PDFInfo
- Publication number
- CN109448065A CN109448065A CN201811181710.3A CN201811181710A CN109448065A CN 109448065 A CN109448065 A CN 109448065A CN 201811181710 A CN201811181710 A CN 201811181710A CN 109448065 A CN109448065 A CN 109448065A
- Authority
- CN
- China
- Prior art keywords
- block
- image
- row
- column
- blocks
- 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
- 238000005259 measurement Methods 0.000 title claims abstract description 51
- 238000000034 method Methods 0.000 title claims abstract description 31
- 230000003044 adaptive effect Effects 0.000 title claims abstract description 11
- 230000006835 compression Effects 0.000 title abstract description 11
- 238000007906 compression Methods 0.000 title abstract description 11
- 239000011159 matrix material Substances 0.000 claims abstract description 56
- 239000013598 vector Substances 0.000 claims abstract description 15
- 230000000903 blocking effect Effects 0.000 claims description 13
- 238000005070 sampling Methods 0.000 claims description 7
- 230000011218 segmentation Effects 0.000 claims description 5
- 238000011176 pooling Methods 0.000 claims description 4
- 238000009499 grossing Methods 0.000 claims description 3
- 238000000638 solvent extraction Methods 0.000 claims description 2
- 238000004364 calculation method Methods 0.000 abstract description 4
- 230000000694 effects Effects 0.000 description 8
- 230000006870 function Effects 0.000 description 4
- 238000001914 filtration Methods 0.000 description 2
- 238000000691 measurement method Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000008602 contraction Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000006837 decompression Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T9/00—Image coding
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of compression sensing methods based on the measurement of gradient block adaptive.This method comprises: using nonoverlapping basic blocks as the division unit of image, the smoothness that each basic blocks are calculated using the gradient both vertically and horizontally of image is divided the image into the non-uniform image block of size according to smoothness and calculates the measured rate of each image block；Each image block is measured respectively according to image block size categorizing selection calculation matrix, obtains measured value；Decoding end introduces vector angle as similitude judgment criteria, is reconstructed using non-local low rank regularization compressed sensing restructing algorithm.The present invention devises a kind of uneven block adaptive measurement compression sensing method, and so that it is decoded image has robustness and obtain better quality reconstruction.
Description
Technical Field
The invention relates to the field of computer image processing, in particular to a compressed sensing method based on gradient block self-adaptive measurement.
Background
Compressed sensing is a new signal processing framework that breaks through the nyquist sampling theorem. Information compression and signal reconstruction are two important components of compressed sensing. The information compression method is mainly divided into two types, and the whole image is compressed and the image is compressed after being partitioned. The whole compressed sensing usually needs to store a larger measurement matrix, occupies a larger memory, and meanwhile, the calculation amount of the whole compressed sensing is also very huge.
Therefore, scientific researchers have proposed a block compression method, which divides an image into small blocks with specified sizes and then compresses each small block of the image by using the same measurement matrix. For example, chinese patent document CN 106301384 a, "a signal reconstruction method based on block compressed sensing": uniformly dividing an original signal into sub-blocks, and filtering after sparse transformation of the sub-blocks; observing the obtained sub-signals to obtain an observation vector; and recovering the sub-signals by using the observation vectors and the measurement matrix, and then linearly combining the sub-signals to obtain a reconstructed signal. The block compressed sensing has the advantages of smaller storage measurement matrix and simpler calculation of a reconstruction algorithm; but this algorithm reconstructs each sub-signal separately and then combines linearly, is not robust and easily causes reconstructed blocking artifacts.
In the document "Compressive sampling based on frequency sampling for remotesensing imaging", in order to improve the block compression sensing reconstruction effect, the measurement rate is allocated according to the significance information of the block, and the reconstruction effect is effectively improved. However, the OMP algorithm is adopted in the reconstruction process, the priori knowledge of the image is not fully utilized, and the set sparsity value has a large influence on the reconstruction effect.
The current block compression sensing algorithm adopts uniform blocking, so that the calculation complexity is reduced by blocking, but the original structure among image blocks is damaged, so that the block effect exists in the block compression sensing algorithm; therefore, many researchers have proposed many filtering methods to smooth the difference between the restored blocks and further remove the blocking effect, but this also results in that the image restoration structure is destroyed while the blocking effect is removed, and the image reconstruction effect is reduced.
The non-local Low-Rank Regularization compressed Sensing reconstruction algorithm proposed in the document Compressive Sensing non-local Low-Rank Regularization fully utilizes the non-local similarity of images to obtain a very good reconstruction effect, but the algorithm adopts integral uniform measurement at a compression end and does not fully utilize measurement information. In order to obtain a more excellent reconstruction effect, the present invention performs reconstruction by using the non-local low rank method at the decoding end.
Disclosure of Invention
In order to solve the above defects in the prior art, solve the problem of block compressed sensing reconstruction and improve the peak signal-to-noise ratio of the reconstructed image, the invention aims to: a compressed sensing method based on gradient block adaptive measurement is provided.
The technical scheme adopted by the invention for solving the technical problems is as follows: in order to reduce the contradiction between the deblocking effect and the preservation of the recovery structure of the blocking method, a non-uniform blocking strategy is provided by combining the image geometric structure, areas with similar structures are divided into one block as much as possible, the whole blocking quantity and the blocking effect of the image are reduced, and the structure loss during the deblocking is further reduced. During measurement, self-adaptive measurement is carried out according to the smoothness of the uneven block, and under the condition that the overall measurement rate is not changed, the measurement rate of a smooth area is reduced and the measurement rate of a complex area is increased according to the smoothness in a self-adaptive mode, so that the utilization rate of effective information is improved.
The method comprises the following steps:
(1) inputting an image to be processed, wherein the number of the image to be processedThe quantization matrix is X ∈ Rm×nM and n are respectively the number of rows and columns of the image matrix;
(2) partitioning the image, dividing the image into d uneven block images based on image gradient, and expanding all image blocks into a column vector xjWhere j ═ {1, 2., d }, are combined into a column vector x ═ x1；x2；...；xd]；
(3) The measurement rate M of each block is assigned according to the required sampling rate M of the image and the smoothness of each blockj,j＝{1,2,...,d}；
(4) Dividing blocks with the same element number into one class, generating Gaussian random orthogonal matrix with the same size, and combining the measurement rate M of each blockjSelecting the front round (M) of the Gaussian random orthogonal matrixj×length(xj) Row as the measurement matrix phij(ii) a Measuring the data to obtain a measured value y ═ Φ x ═ diag (Φ)1,φ2,...,φd)[x1；x2；...；xd]；
(5) Reconstructing an original signal by using a measured value, a block measurement rate, a row segmentation position and a measurement matrix, solving by using a non-local low-rank regularization compressed sensing reconstruction algorithm, and outputting a reconstructed image;
xjexpressing a column vector formed by expanding the partitioned image block matrix according to columns, wherein j is the sequence of the image block and is d blocks;
φjdenotes xjThe measuring matrix selects the row number according to the sampling rate, and the column number is determined by the size of the image block;
yjgenerating a column vector measurement value for each image block;
the blocking algorithm in the step (2) is as follows:
step 1: calculating X ∈ R of imagem×nVertical gradient matrix DvAnd a horizontal direction gradient matrix Dh：
Where X (i, j) denotes the pixel values in i rows and j columns of the image matrix X, Dv(i,j)、Dv(i, j) represent the vertical gradient and the horizontal gradient in the image matrix X at i rows and j columns, respectively.
Step 2: the image X is divided into basic blocks of r × r which are not overlapped with each other, and r is a multiple of 2, and if r × r is less than r × r, the image is filled with 0.
And step 3: respectively to the gradient matrix D in the vertical directionvAnd a horizontal direction gradient matrix DhPooling operation is performed by using basic blocks as units to respectively obtain vertical smooth matrixesAnd a horizontal direction smoothing matrixAnd calculating smoothness representing the vertical and horizontal directions of each basic block according to the gradient in the vertical direction and the gradient in the horizontal direction respectively:
wherein S isv(p,q)、Sh(p, q) respectively representing the vertical direction smoothness and the horizontal direction smoothness of the p-th row and q-th column basic block in units of basic blocks; e is constant, avoiding denominator being zero, and when e is taken as 1, Sv(p,q),Sh(p,q)∈(1,255]Larger values represent better smoothness.
And 4, step 4: will SvAre divided into two classes, a threshold value T is setv，Sv(i,j)＜TvIs a non-flat slide block, Sv(i,j)≥TvFor a flat slide block, note Nv(i) Calculating line segmentation parameters for the number of the ith line of smooth blocks:
the ratio of the number of sliders in the ith row is expressed, and the ratio of the number of sliders is used as a reference value for row division because the image structure is regional.
And 5: according to the vertical smoothness matrix SvThe basic block image is divided into line blocks from top to bottom, RvStore the line index, length (R) of the last line of basic blocks in a line blockv) The final row block number.
5.1: initial i ═ 1, j ═ 1, Rv(j)＝1，σvIs constant and represents the line block division threshold.
5.2: i +1, judge | Lv(i)-Lv(i-1)|＜σvIf yes, indicating that the smoothness of the ith basic block row and the ith-1 basic block row in the vertical direction is within the threshold value range, classifying the ith basic block row into the same row block containing the ith-1 basic block row, wherein R isv(j) I; otherwise, indicating that the vertical smoothness of the two basic block rows is not within the threshold value, dividing the ith basic block row into the next row of blocks, wherein j is j +1, and R isv(j)＝i。
5.3: judgment ofIf yes, the line block division is finished, step 6 is executed, and if not, 5.2 is executed.
Step 6: for each divided row of blocks, according to a horizontal smoothness matrix ShThe row and column division proceeds from left to right, and the division method is similar to the row division method.
6.1: initial j is 0, Rv(0)＝0；
6.2: j equals j +1, extracting ShJ-th row block s ofh，sh＝Sh(Rv(j-1)+1:Rv(j) In pair with shAre divided into two classes, a threshold value T is seth，sh(i,j)＜ThIs a non-flat slide block sh(i,j)≥ThFor a flat slide block, note Nh(k) Denotes shCalculating the number of horizontal sliding blocks in the kth column, and calculating column division parameters:
wherein L ish(k) Denotes shThe ratio of the k-th row of sliders, RvStores a row index of a last row basic block of the division row block,
Rv(j)-Rv(j-1) represents shNumber of basic blocks in each column, the value of which is equal to shThe number of elements in each column,.
6.3: to shColumn division, Rh(j, k) storing the last column index of the kth column block dividing the jth row block. length (R)h(j,: is) is shThe number of column blocks is different from the smoothness of each row block, so that the number of divided column blocks is different from each row block.
6.4: initial i ═ 1, k ═ 1, Rh(j,k)＝1，σhThe column block threshold is divided to be constant.
6.5: i +1, judge | Lh(k)-Lh(k-1)|＜σhIf yes, the horizontal smoothness of the k-th basic block column and the k-1 th basic block column in the j-th row block is within the threshold value rangeThen the kth base block column is placed in the same column of blocks that contains the kth-1 base block column, where R ish(j, k) ═ i; otherwise, the horizontal direction smoothness of the two basic block columns is not within the threshold value, the k-th basic block column is divided into the blocks of the next column, wherein k is k +1, and R ish(j,k)＝i。
6.6: judgment ofIf not, indicating that the column division of the jth row block is not finished, and executing 6.5; if yes, judging j ═ length (R)v) If not, executing step 6.2, if yes, completing all the blocks and saving Rh(j,k)。
The measurement rate in the step (3) adopts a self-adaptive distribution algorithm, and the algorithm is as follows:
step 1: input image global measurement rate M, calculating image block xj1, the smoothness sum of all basic blocks in q:
step 2: the measurement rate is assigned according to sub-block smoothness:
wherein M is0Block lowest measurement rate > 0.
The step (4) adopts a non-local low-rank regularization compressed sensing reconstruction algorithm to reconstruct, and the column vectors are recombined into images for processing during reconstruction;
due to the adoption of the technical scheme, the invention has the beneficial effects that: by adopting the block compressed sensing image reconstruction algorithm, the blocking effect caused by block compressed sensing is effectively reduced; compared with the existing self-adaptive block measurement method, the method has higher peak signal-to-noise ratio and better image visual effect.
Drawings
FIG. 1 is a general flow chart of the algorithm of a compressed sensing method based on gradient block adaptive measurement according to the present invention;
FIG. 2 is a flow chart of the gradient non-uniform chunking based algorithm of the present invention;
Detailed Description
The following detailed description is made with reference to the accompanying drawings and an exemplary embodiment for a gradient block adaptive measurement-based compressed sensing method according to the present invention, where the algorithm specifically includes the following parts:
the compression end compresses the image by using a gradient-nonuniform-block-based self-adaptive measurement method, and the method comprises the following steps of:
inputting an image to be processed X e Rm×nCalculating the gradients of the image in the horizontal direction and the vertical direction; performing pooling dimensionality reduction on the image gradient according to the size of the basic block to obtain the smoothness of the image basic block in the horizontal direction and the smoothness of the image basic block in the vertical direction; the image is divided into row blocks from top to bottom according to the smoothness in the vertical direction, and then each row block is divided into column blocks from left to right according to the smoothness in the horizontal direction, so that uneven blocking is completed. The method comprises the following specific steps:
step 1: calculating X ∈ R of imagem×nVertical gradient matrix DvAnd a horizontal direction gradient matrix Dh：
Where X (i, j) denotes the pixel values in i rows and j columns of the image matrix X, Dv(i,j)、Dv(i, j) represent the vertical gradient and the horizontal gradient in the image matrix X at i rows and j columns, respectively.
Step 2: the image X is divided into basic blocks of r × r which are not overlapped with each other, and r is a multiple of 2, and if r × r is less than r × r, the image is filled with 0.
And step 3: respectively to the gradient matrix D in the vertical directionvAnd a horizontal direction gradient matrix DhPooling operation is performed by using basic blocks as units to respectively obtain vertical smooth matrixesAnd a horizontal direction smoothing matrixAnd calculating smoothness representing the vertical and horizontal directions of each basic block according to the gradient in the vertical direction and the gradient in the horizontal direction respectively:
wherein S isv(p,q)、Sh(p, q) respectively representing the vertical direction smoothness and the horizontal direction smoothness of the p-th row and q-th column basic block in units of basic blocks; e is constant, avoiding denominator being zero, and when e is taken as 1, Sv(p,q),Sh(p,q)∈(1,255]Larger values represent better smoothness.
And 4, step 4: will SvThe elements of each row of (1) are divided into two classesSetting a threshold value Tv，Sv(i,j)＜TvIs a non-flat slide block, Sv(i,j)≥TvFor a flat slide block, note Nv(i) Calculating line segmentation parameters for the number of the ith line of smooth blocks:
the ratio of the number of sliders in the ith row is expressed, and the ratio of the number of sliders is used as a reference value for row division because the image structure is regional.
And 5: according to the vertical smoothness matrix SvThe basic block image is divided into line blocks from top to bottom, RvStore the line index, length (R) of the last line of basic blocks in a line blockv) The final row block number.
5.1: initial i ═ 1, j ═ 1, Rv(j)＝1，σvIs constant and represents the line block division threshold.
5.2: i +1, judge | Lv(i)-Lv(i-1)|＜σvIf yes, indicating that the smoothness of the ith basic block row and the ith-1 basic block row in the vertical direction is within the threshold value range, classifying the ith basic block row into the same row block containing the ith-1 basic block row, wherein R isv(j) I; otherwise, indicating that the vertical smoothness of the two basic block rows is not within the threshold value, dividing the ith basic block row into the next row of blocks, wherein j is j +1, and R isv(j)＝i。
5.3: judgment ofIf yes, the line block division is finished, step 6 is executed, and if not, 5.2 is executed.
Step 6: for each divided row of blocks, according to a horizontal smoothness matrix ShThe row and column division proceeds from left to right, and the division method is similar to the row division method.
6.1: initial j is 0, Rv(0)＝0；
6.2: j equals j +1, extracting ShJ-th row block s ofh，sh＝Sh(Rv(j-1)+1:Rv(j) In pair with shAre divided into two classes, a threshold value T is seth，sh(i,j)＜ThIs a non-flat slide block sh(i,j)≥ThFor a flat slide block, note Nh(k) Denotes shCalculating the number of horizontal sliding blocks in the kth column, and calculating column division parameters:
wherein L ish(k) Denotes shThe ratio of the k-th row of sliders, RvStores a row index of a last row basic block of the division row block,
Rv(j)-Rv(j-1) represents shNumber of basic blocks in each column, the value of which is equal to shThe number of elements in each column,.
6.3: to shColumn division, Rh(j, k) storing the last column index of the kth column block dividing the jth row block. length (R)h(j,: is) is shThe number of column blocks is different from the smoothness of each row block, so that the number of divided column blocks is different from each row block.
6.4: initial i ═ 1, k ═ 1, Rh(j,k)＝1，σhThe column block threshold is divided to be constant.
6.5: i +1, judge | Lh(k)-Lh(k-1)|＜σhIf yes, indicating that the smoothness of the k-th basic block column in the j-th row block and the smoothness of the k-1-th basic block column in the horizontal direction are within the threshold value range, classifying the k-th basic block column into the same column block containing the k-1-th basic block column, wherein R ish(j, k) ═ i; otherwise, the horizontal direction smoothness of the two basic block columns is not in the threshold value, and the k basic block column is processedDivided into the next column of blocks, where k is k +1, Rh(j,k)＝i。
6.6: judgment ofIf not, indicating that the column division of the jth row block is not finished, and executing 6.5; if yes, judging j ═ length (R)v) If not, executing step 6.2, if yes, completing all the blocks and saving Rh(j,k)。
The measurement rate M for each block is then assigned according to the sampling rate M required to compress the image and the smoothness of each blockjJ ═ 1,2,.. d }, the concrete steps are as follows:
step 1: input image global measurement rate M, calculating image block xj1, the smoothness sum of all basic blocks in q:
step 2: the measurement rate is assigned according to sub-block smoothness:
wherein M is0Block lowest measurement rate > 0.
Finally, the blocks with the same element number are divided into one class, the Gaussian random orthogonal matrix with the same size is generated, and the measurement rate M of each block is combinedjSelecting the front round (M) of the Gaussian random orthogonal matrixj×length(xj) Row as the measurement matrix phij(ii) a Measuring the data to obtain a measured value y ═ Φ x ═ diag (Φ)1,φ2,...,φd)[x1；x2；...；xd]，；
The decompression end adopts a non-local low-rank regularization compressed sensing reconstruction algorithm, and the specific steps are as follows:
step 1: obtaining an initial estimated image x by using a DCT soft threshold algorithm according to the measured value, the block measurement rate, the line division position and the measurement matrix(0)；
Step 2: initialization, k being 1, x(k)＝x(0)And k represents the number of iterations. X is divided by step 51Dividing into 6 × 6 overlapped image reference blocks, and dividing into image x according to each reference block1The 44 most similar blocks are searched. The invention adopts the cosine included angle function of the vector to judge the similarity:
wherein,representing an image x(k)The column vector of the ith reference block of (1),representing an image x(k)Column vector of image block of medium size 6 × 6, c denotes image blockAnd reference blockThe larger the value of (a), the larger the similarity, and the maximum value of 1. Forming a similar block matrix X by 45 similar block column vectors including the reference blockiWherein X isi＝Rix＝[Ri0x,Ri1x,...,Riqx]R denotes an extraction block operation, and q is 45.
And step 3: modeling, similar Block matrix XiThe low rank model of (c) is:
wherein,is the variance of gaussian noise.
Taking the low-rank model as a regular term of the reconstruction model as follows:
wherein η and lambda are regularization parameters and are set as constants during reconstruction, and L (L)iEpsilon) represents a rank substitution function, a kernel function is adopted as the rank substitution function,
and (3) converting into:
and 4, step 4: solving similar block matrix XiLow rank matrix L ofiThe singular value contraction algorithm is adopted to obtain:
wherein,σjto representJ-th singular value of (U Σ V)TIs thatSingular value decomposition matrix of (x)+＝max{x,0}。
Step 5, solving the reconstructed imageJudging a termination condition, if the output reconstructed image is satisfied, otherwise, when mod (k, T) is 0, re-searching each reference block for the most similar 45 similar blocks to form a similar block matrix XiGo to step 3. Wherein,representing an operation of restoring the similar blocks in the ith group of similar blocks to the original image position,
it is to be understood that the above description is only one specific embodiment of the invention and that the invention is not limited to the specific constructions shown and described, since the claims are intended to cover all modifications that are within the true spirit and scope of the invention.
Claims (4)
1. A compressed sensing method based on gradient block adaptive measurement is characterized by comprising the following steps:
the method comprises the following steps:
(1) inputting an image to be processed, wherein the digital matrix of the image to be processed is X epsilon Rm×nM and n are respectively the number of rows and columns of the image matrix;
(2) partitioning the image, dividing the image into d uneven image blocks based on the image gradient, and expanding all the image blocks into a column vector xjWherein j ═ {1, 2.. multidata, d }, are combined into a columnVector x ═ x1；x2；...；xd]；
(3) The measurement rate M of each block is assigned according to the sampling rate M required for compressing the image and the smoothness of each blockj,j＝{1,2,...,d}；
(4) Dividing blocks with the same element number into one class, generating Gaussian random orthogonal matrix with the same size, and combining the measurement rate M of each blockjSelecting the front round (M) of the Gaussian random orthogonal matrixj×length(xj) Row as the measurement matrix phij(ii) a Measuring the data to obtain a measured value y ═ Φ x ═ diag (Φ)1,φ2,...,φd)[x1；x2；...；xd]；
(5) And reconstructing the original signal by using the measured value, the block measurement rate, the row segmentation position and the measurement matrix, solving by using a non-local low-rank regularization compressed sensing reconstruction algorithm, and outputting a reconstructed image.
2. The compressed sensing method based on gradient block adaptive measurement as claimed in claim 1, wherein the blocking algorithm in step (2) is as follows:
step 1: calculating X ∈ R of imagem×nVertical gradient matrix DvAnd a horizontal direction gradient matrix Dh：
Where X (i, j) denotes the pixel values in i rows and j columns of the image matrix X, Dv(i,j)、Dv(i, j) respectively representing the vertical direction gradient and the horizontal direction gradient of the image matrix X in i rows and j columns;
step 2: dividing the image X into non-overlapping r × r basic blocks, wherein r is a multiple of 2, and if r × r is less than r × r, the r × r basic blocks are filled with 0;
and step 3: are respectively provided withFor vertical direction gradient matrix DvAnd a horizontal direction gradient matrix DhPooling operation is performed by using basic blocks as units to respectively obtain vertical smooth matrixesAnd a horizontal direction smoothing matrixAnd calculating smoothness representing the vertical and horizontal directions of each basic block according to the gradient in the vertical direction and the gradient in the horizontal direction respectively:
wherein S isv(p,q)、Sh(p, q) respectively representing the vertical direction smoothness and the horizontal direction smoothness of the p-th row and q-th column basic block in units of basic blocks; e is constant, avoiding denominator being zero, and when e is taken as 1, Sv(p,q),Sh(p,q)∈(1,255]Larger values represent better smoothness;
and 4, step 4: will SvAre divided into two classes, a threshold value T is setv，Sv(i,j)＜TvIs a non-flat slide block, Sv(i,j)≥TvFor a flat slide block, note Nv(i) Calculating line segmentation parameters for the number of the ith line of smooth blocks:
the proportion of the ith row of smooth blocks is shown, and the number proportion of the smooth blocks is adopted as a reference value for row division because the image structure has regionality;
and 5: according to the vertical smoothness matrix SvFor line-blocking the basic block image from top to bottomSplitting, RvStore the line index, length (R) of the last line of basic blocks in a line blockv) The number of the final line blocks;
5.1: initial i ═ 1, j ═ 1, Rv(j)＝1，σvIs a constant, representing a row block division threshold;
5.2: i +1, judge | Lv(i)-Lv(i-1)|＜σvIf yes, indicating that the smoothness of the ith basic block row and the ith-1 basic block row in the vertical direction is within the threshold value range, classifying the ith basic block row into the same row block containing the ith-1 basic block row, wherein R isv(j) I; otherwise, indicating that the vertical smoothness of the two basic block rows is not within the threshold value, dividing the ith basic block row into the next row of blocks, wherein j is j +1, and R isv(j)＝i；
5.3: judgment ofIf yes, the line block division is finished, step 6 is executed, and if not, 5.2 is executed;
step 6: for each divided row of blocks, according to a horizontal smoothness matrix ShPerforming row-column division from left to right, wherein the division method is similar to the row division method;
6.1: initial j is 0, Rv(0)＝0；
6.2: j equals j +1, extracting ShJ-th row block s ofh，sh＝Sh(Rv(j-1)+1:Rv(j) In pair with shAre divided into two classes, a threshold value T is seth，sh(i,j)＜ThIs a non-flat slide block sh(i,j)≥ThFor a flat slide block, note Nh(k) Denotes shCalculating the number of horizontal sliding blocks in the kth column, and calculating column division parameters:
wherein L ish(k) Denotes shThe ratio of the k-th row of sliders, RvStoring a row index of a last row basic block of a split row block，Rv(j)-Rv(j-1) represents shNumber of basic blocks in each column, the value of which is equal to shThe number of elements in each column;
6.3: to shColumn division, Rh(j, k) storing the last column index of the kth column block dividing the jth row block. length (R)h(j,: is) is shThe number of column blocks, and the quantity of the divided column blocks of each row block is different because the smoothness of each row block is different;
6.4: initial i ═ 1, k ═ 1, Rh(j,k)＝1，σhDividing a column block threshold value into constants;
6.5: i +1, judge | Lh(k)-Lh(k-1)|＜σhIf yes, indicating that the smoothness of the k-th basic block column in the j-th row block and the smoothness of the k-1-th basic block column in the horizontal direction are within the threshold value range, classifying the k-th basic block column into the same column block containing the k-1-th basic block column, wherein R ish(j, k) ═ i; otherwise, the horizontal direction smoothness of the two basic block columns is not within the threshold value, the k-th basic block column is divided into the blocks of the next column, wherein k is k +1, and R ish(j,k)＝i；
6.6: judgment ofIf not, indicating that the column division of the jth row block is not finished, and executing 6.5; if yes, judging j ═ length (R)v) If not, executing step 6.2, if yes, completing all the blocks and saving Rh(j,k)。
3. The compressed sensing method based on gradient block adaptive measurement as claimed in claim 1, wherein the measurement rate in step (3) is adaptive distribution algorithm, the algorithm is as follows:
step 1: input image global measurement rate M, calculating image block xj1, the smoothness sum of all basic blocks in q:
step 2: the measurement rate is assigned according to sub-block smoothness:
wherein M is0Block lowest measurement rate > 0.
4. The gradient block adaptive measurement-based compressed sensing method according to claim 1, wherein the step (5) adopts a similar block judgment standard in a non-local low-rank regularization compressed sensing reconstruction algorithm:
wherein,representing an image x(k)The column vector of the ith reference block of (1),representing an image x(k)Column vector of image block of medium size 6 × 6, c denotes image blockAnd reference blockThe larger the value of (a), the larger the similarity, and the maximum value of 1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811181710.3A CN109448065B (en) | 2018-10-11 | 2018-10-11 | Compressed sensing method based on gradient blocking self-adaptive measurement |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811181710.3A CN109448065B (en) | 2018-10-11 | 2018-10-11 | Compressed sensing method based on gradient blocking self-adaptive measurement |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109448065A true CN109448065A (en) | 2019-03-08 |
| CN109448065B CN109448065B (en) | 2023-07-25 |
Family
ID=65545399
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811181710.3A Active CN109448065B (en) | 2018-10-11 | 2018-10-11 | Compressed sensing method based on gradient blocking self-adaptive measurement |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109448065B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110418137A (en) * | 2019-07-31 | 2019-11-05 | 东华大学 | A kind of rest block collection measured rate regulation method for intersecting subset guiding |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070248164A1 (en) * | 2006-04-07 | 2007-10-25 | Microsoft Corporation | Quantization adjustment based on texture level |
| CN103440675A (en) * | 2013-07-30 | 2013-12-11 | 湖北工业大学 | Overall situation reconstitution optimization model construction method for image block compressed sensing |
| CN107154061A (en) * | 2017-05-09 | 2017-09-12 | 北京航宇天穹科技有限公司 | The regularization coding/decoding method that a kind of splits' positions are perceived |
-
2018
- 2018-10-11 CN CN201811181710.3A patent/CN109448065B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070248164A1 (en) * | 2006-04-07 | 2007-10-25 | Microsoft Corporation | Quantization adjustment based on texture level |
| CN103440675A (en) * | 2013-07-30 | 2013-12-11 | 湖北工业大学 | Overall situation reconstitution optimization model construction method for image block compressed sensing |
| CN107154061A (en) * | 2017-05-09 | 2017-09-12 | 北京航宇天穹科技有限公司 | The regularization coding/decoding method that a kind of splits' positions are perceived |
Non-Patent Citations (2)
| Title |
|---|
| SIWANG ZHOU 等: "Block compressed sampling of image signals by saliency based adaptive partitioning", 《MULTIMEDIA TOOLS AND APPLICATIONS》 * |
| 杜秀丽等: "基于灰度共生矩阵的图像自适应分块压缩感知方法", 《计算机科学》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110418137A (en) * | 2019-07-31 | 2019-11-05 | 东华大学 | A kind of rest block collection measured rate regulation method for intersecting subset guiding |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109448065B (en) | 2023-07-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105513026B (en) | One kind being based on the non local similar compressed sensing reconstructing method of image | |
| CN102722896B (en) | Adaptive compressed sensing-based non-local reconstruction method for natural image | |
| CN107730451B (en) | Compressed sensing reconstruction method and system based on depth residual error network | |
| CN110490832A (en) | A kind of MR image reconstruction method based on regularization depth image transcendental method | |
| CN106952317B (en) | Hyperspectral image reconstruction method based on structure sparsity | |
| CN110827198B (en) | Multi-camera panoramic image construction method based on compressed sensing and super-resolution reconstruction | |
| CN102332153B (en) | Kernel regression-based image compression sensing reconstruction method | |
| CN104199627B (en) | Gradable video encoding system based on multiple dimensioned online dictionary learning | |
| CN113808032A (en) | Multi-stage progressive image denoising algorithm | |
| CN110728728B (en) | Compressed sensing network image reconstruction method based on non-local regularization | |
| CN109102461A (en) | Image reconstructing method, device, equipment and the medium of low sampling splits' positions perception | |
| Shi et al. | Multi-scale deep networks for image compressed sensing | |
| CN105590296B (en) | A kind of single-frame images Super-Resolution method based on doubledictionary study | |
| CN112218094A (en) | JPEG image decompression effect removing method based on DCT coefficient prediction | |
| CN107154061B (en) | Regularized decoding method for block compressed sensing | |
| CN113284202B (en) | Image compression sensing method of scalable network based on content self-adaption | |
| CN104952051A (en) | Low-rank image restoration method based on Gaussian mixture model | |
| CN106558020B (en) | A kind of image rebuilding method and system based on network image block retrieval | |
| CN109448065A (en) | A kind of compression sensing method based on the measurement of gradient block adaptive | |
| CN112150360A (en) | IVUS image super-resolution reconstruction method based on dense residual error network | |
| CN115131226B (en) | Image restoration method based on wavelet tensor low-rank regularization | |
| CN111510719B (en) | Distributed compressed sensing coding and decoding method for video | |
| Suryanarayana et al. | Deep Learned Singular Residual Network for Super Resolution Reconstruction. | |
| Li et al. | Compression artifact removal with stacked multi-context channel-wise attention network | |
| Chen et al. | Rapid-INR: Storage Efficient CPU-free DNN Training Using Implicit Neural Representation |
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 |