CN108093266A - Utilize a group compression of images sensing reconstructing system and method for normalization rarefaction representation - Google Patents

Utilize a group compression of images sensing reconstructing system and method for normalization rarefaction representation Download PDF

Info

Publication number
CN108093266A
CN108093266A CN201711338419.8A CN201711338419A CN108093266A CN 108093266 A CN108093266 A CN 108093266A CN 201711338419 A CN201711338419 A CN 201711338419A CN 108093266 A CN108093266 A CN 108093266A
Authority
CN
China
Prior art keywords
image
module
matrix
unit
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
Application number
CN201711338419.8A
Other languages
Chinese (zh)
Other versions
CN108093266B (en
Inventor
熊承义
高志荣
龚忠毅
张梦杰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
South Central Minzu University
Original Assignee
South Central University for Nationalities
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by South Central University for Nationalities filed Critical South Central University for Nationalities
Priority to CN201711338419.8A priority Critical patent/CN108093266B/en
Publication of CN108093266A publication Critical patent/CN108093266A/en
Application granted granted Critical
Publication of CN108093266B publication Critical patent/CN108093266B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/90—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using coding techniques not provided for in groups H04N19/10-H04N19/85, e.g. fractals
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
    • H04N19/102—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
    • H04N19/117—Filters, e.g. for pre-processing or post-processing
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
    • H04N19/102—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
    • H04N19/132—Sampling, masking or truncation of coding units, e.g. adaptive resampling, frame skipping, frame interpolation or high-frequency transform coefficient masking
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
    • H04N19/134—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or criterion affecting or controlling the adaptive coding
    • H04N19/154—Measured or subjectively estimated visual quality after decoding, e.g. measurement of distortion
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
    • H04N19/169—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the coding unit, i.e. the structural portion or semantic portion of the video signal being the object or the subject of the adaptive coding
    • H04N19/17—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the coding unit, i.e. the structural portion or semantic portion of the video signal being the object or the subject of the adaptive coding the unit being an image region, e.g. an object
    • H04N19/176—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the coding unit, i.e. the structural portion or semantic portion of the video signal being the object or the subject of the adaptive coding the unit being an image region, e.g. an object the region being a block, e.g. a macroblock
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/60—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding
    • H04N19/63—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding using sub-band based transform, e.g. wavelets
    • H04N19/635—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding using sub-band based transform, e.g. wavelets characterised by filter definition or implementation details
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/70—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals characterised by syntax aspects related to video coding, e.g. related to compression standards

Landscapes

  • Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Image Processing (AREA)

Abstract

The invention discloses a kind of compression of images sensing reconstructing system and method using group normalization rarefaction representation, are related to image-recovery technique field.The system includes initialization module (1), routing selecting module (2), regularization mean square error minimum module (3) and image filtering processing module (4)；Initialization module (1), routing selecting module (2), regularization mean square error minimum module (3) and image filtering processing module (4) interact successively, image filtering processing module (4) and routing selecting module (2) interaction.The present invention can improve the recovery effects at image Weak target and texture and edge, effectively promote the reconstruction quality of compressed sensing image；Suitable for compressing imaging applications.

Description

Utilize a group compression of images sensing reconstructing system and method for normalization rarefaction representation
Technical field
The present invention relates to image-recovery technique field more particularly to a kind of image pressures using group normalization rarefaction representation Contracting sensing reconstructing system and method.
Background technology
As a kind of new signal sampling theory proposed in recent years, compressed sensing (Compressive Sensing, CS) Sampling theory breaches the constraint of conventional Nyquist sampling theory, it can be achieved that the accurate reconstruction sampled to sparse signal dimensionality reduction. The theory obtained since the proposition extensive concern in the fields such as sensing, image procossing, wireless communication and answered With.Compression of images perception is with a wide range of applications in fields such as remote sensing images, medical imagings, and the height of compressed sensing image Quality reconstruction is the key point of its successful application.
In order to realize the accurate reconstruction of compressed sensing signal, reconstruction property is more preferably promoted, finds and is suitble to the dilute of sampled signal It dredges and represents particularly significant；And it is reconstructed for the compressed sensing of picture signal, the Local Phase that picture signal how to be made full use of to have The non local similar rarefaction representation ability more preferably to be promoted to image of Sihe, so as to further effectively promote image reconstruction quality, It is the important directions that area research persons pay close attention to always in recent years, such as the method based on total variation【[1]Li C,Yin W, Zhang Y.TVAL3:TV minimization by augmented Lagrangian and alternating direction algorithms[EB/OL].2013. http://www.caam.rice.edu/optimization/L1/ TVAL3/.】, based on the method more assumed【[2]CHEN C,TRAMEL E W.Compressed-sensing recovery of images and video using multi-hypothesis predictions[C]//In Proc.45th Asilomar Conf.Signals,Syst.,Comput. Pacific Grove,CA,USA,2011.1193-1198.】, based on non local phase Like the method for (NLS)【[3]Zhang J,Zhao D,Gao W.Group-based sparse representation for image restoration[J].IEEE Transactions on Image Processing,2014,23(8):3336- 3351.[4] Shen Y F,Zhu Z M,Zhang Y D.Compressed sensing image reconstruction algorithm based on rank minimization[J].Acta Electronica Sinica,2016,44(3): 572-579.[5]Li J, Gao Z,Xiong C,ZhouC.Image compressive sensing recovery based on weighted structure group sparse representation.Journal on Communications, 2017, 38(2):2017041.】, etc..
However, previous methods mostly simply directly using picture signal by certain conversion after it is openness come constrain weight Thus structure can cause detailed information small and important present in image to be lost, so as to be difficult to obtain satisfied quality reconstruction.
The content of the invention
The purpose of the invention is to overcome the shortcomings of above-mentioned background technology, an a kind of utilization group normalization rarefaction representation is provided Compressed sensing image reconstruction system and its method, can preferably retain the details such as image Weak target and texture and edge letter Breath, so as to effectively promote the quality of reconstructed image.
The present invention technical thought be：
The initial estimate of compressed sensing image reconstruction is obtained using compression of images sensing reconstructing method first；Then again The mean normalization is gone to represent the priori with approximate consistent statistical distribution using the different sub-band coefficient of block image, by more Secondary iteration goes average soft-threshold de-noising to handle, effectively promotes the reconstruct matter of compressed sensing image using image block group PCA transform domains Amount.
First, a group compressed sensing image reconstruction system (abbreviation system) for normalization rarefaction representation is utilized
The system is included at initialization module, routing selecting module, regularization mean square error minimum module and image filtering Manage module；
Initialization module, routing selecting module, regularization mean square error minimum module and image filtering processing module are successively Interaction, image filtering processing module and routing selecting module interaction.
2nd, a group compressed sensing image reconstructing method (abbreviation method) for normalization rarefaction representation is utilized
This method comprises the following steps：
1. initialization module receives the compression sampling value y of input picture, and using the reconstructing method of compressed sensing image, it is raw Into the initial estimate z of compressed sensing image reconstruction, routing selecting module is sent into, the graceful distance a=0 of initial setting up Donald Bragg, is changed Generation number t=0, maximum iteration J, regularisation parameter λ and μ；
2. the estimation for the compressed sensing image reconstruction that regularization mean square error minimum module is inputted according to routing selecting module The compression sampling value y of value z and input picture, pass through solutionObtain reconstructed image Once newer estimate x, be sent into image filtering processing module；Φ is calculation matrix, is usually chosen for random matrix, and a is The graceful distance of Donald Bragg；
3. the reconstructed image that image filtering processing module inputs regularization mean square error minimum module is once newer Estimate x carries out based on coefficient in transform domain grade average soft-threshold de-noising being gone to handle, and obtains the secondary of reconstructed image and newer estimates Evaluation z；
4. update the graceful distance a of Donald Bragg：A=a- (x-z), update iterations t=t+1；If iterations t is little In the maximum iteration J of setting, then back to step 2.；Otherwise step is performed 5.；
5. terminating computing, the reconstructed image z finally obtained is exported.
Working mechanism：
The realization of the present invention is totally divided into two stages：
First stage obtains the initial estimate of compressed sensing image reconstruction using compression of images sensing reconstructing method；
Second stage, using image with block image different sub-band coefficient go mean normalization represent have closely Like the priori of consistent statistical distribution, using a group optimization for normalization rarefaction representation regularization, compression is promoted by multiple iteration The quality of perceptual image reconstruct, iteration therein include two processes：First process is adopted according to the compression of known image Sample value once updates the estimate of reconstructed image using regularization mean square error minimum method；Second process is root According to the estimate of once newer reconstructed image, average soft-threshold de-noising method is gone to reconstruct using image similar block group transform domain The estimate of image carries out secondary update.
Compared with prior art, the present invention has following advantages and good effect：
The present invention can improve the recovery effects at image Weak target and texture and edge, effectively promote compressed sensing image Reconstruction quality；Suitable for compressing imaging applications.
Description of the drawings
Fig. 1 is the block diagram of the system；
In figure：
1-initialization module；
2-routing selecting module；
3-regularization mean square error minimum module；
4-image filtering processing module,
401-image overlap partition unit,
402-image similar block group generation unit,
403-transform domain removes average soft-threshold de-noising unit,
404-image block pixel domain is averaging unit.
Fig. 2 .1 are experiment original images,
Fig. 2 .2 are the reconstructed images using document [2] method,
Fig. 2 .3 are the reconstructed images using document [3] method,
Fig. 2 .4 are the reconstructed images using the method for the present invention.
Specific embodiment
It is described in detail with reference to the accompanying drawings and examples：
First, system
1st, it is overall
Such as Fig. 1, the system includes initialization module 1, routing selecting module 2, regularization mean square error minimum module 3 and figure As filtering process module 4；
Initialization module 1, routing selecting module 2, regularization mean square error minimum module 3 and image filtering processing module 4 It interacts successively, image filtering processing module 4 and routing selecting module 2 interact.
In detail：There are two input terminal, an output terminal, regularization mean square error minimum modules 3 for routing selecting module 2 There are 2 input terminals, 1 output terminal；One input terminal of routing selecting module 2 is interacted with the output terminal of initialization module 1, routing The output terminal that another input terminal of selecting module 2 is averaging unit 404 with image block pixel domain interacts, regularization mean square error One input terminal of poor minimum module 3 is interacted with system input, another input terminal and the output terminal of 2 module of Route Selection are handed over Mutually, the output terminal of regularization mean square error minimum module 3 is interacted with the input terminal of image overlap partition unit 401, image overlapping The output terminal of blocking unit 401 is interacted with the input terminal of image similar block group generation unit 402, image similar block group generation unit 402 output terminal goes the input terminal of average soft-threshold de-noising unit 403 to interact with transform domain, and transform domain removes average soft-threshold de-noising The input terminal that the output terminal of unit 403 is averaging unit 404 with image block pixel domain interacts, and image block pixel domain is averaging unit 404 output terminal is the output terminal of compression of images sensing reconstructing system.
2nd, function module
1) initialization module 1
Initialization module 1 receives the compression sampling value of input picture, and using the method for conventional image compression sensing reconstructing Such as the method (more other similar approach can also be used) of more hypothesis of document [3] proposition, compressed sensing image weight is obtained The initial estimate of structure is sent into routing selecting module 2.
2) routing selecting module 2
Routing selecting module 2 exports the initial estimate of the compressed sensing image reconstruction of the feeding of initialization module 1 for the first time, The newer estimate that image block pixel domain is averaging the compressed sensing image reconstruction that unit 404 is sent into is exported afterwards.
3) regularization mean square error minimum module 3
Regularization mean square error minimum module 3 according to routing selecting module 2 be sent into image reconstruction estimate and according to The compression sampling value of the image of input obtains the once newer estimate of reconstructed image.Especially by functionIt differentiates on variable x and gradient descent method iterative solution is obtained or utilized equal to zero.
4) image filtering processing module 4
Image filtering processing module 4 includes image overlap partition unit 401 interactive successively, image similar block group generation list Member 402, transform domain go average soft-threshold de-noising unit 403 and image block pixel domain to be averaging unit 404；
(1) image overlap partition unit 401
Image overlap partition unit 401 carries out overlap partition to the image that regularization mean square error minimum module 3 exports, and obtains To multiple image blocks there are pixel overlapping, image similar block group generation unit 402 is sent into；
(2) image similar block group generation unit 402
Each image block that image similar block group generation unit 402 is sent into image overlap partition Unit 401, where it In the regulation contiguous range of image, one group of similar image block is searched out, by generation after this group of similar image block vector quantization and the figure As the corresponding data matrix of block, it is sent into transform domain and removes average soft-threshold de-noising unit 403；
(3) transform domain removes average soft-threshold de-noising unit 403
Transform domain goes average soft-threshold de-noising unit 403 to generate each data that Unit 402 are sent into image similar block group Matrix generates PCA conversion dictionaries according to its covariance matrix first, and then conversion dictionary is multiplied with data matrix is converted Next domain coefficient matrix asks for average and variance that coefficient in transform domain matrix corresponds to each subband, and further to transform domain Each coefficient according to its affiliated subband difference carry out average soft-threshold shrink process respectively, finally by inverse transformation dictionary with As a result the coefficient in transform domain matrix multiple of soft-threshold shrink process is sent into image block pixel domain and is averaging unit 404；
(4) image block pixel domain is averaging unit 404
Image block pixel domain is averaging the data square that unit 404 removes transform domain average soft-threshold de-noising unit 403 to be sent into Battle array using carrying out seeking weighted average to respective pixel in pixel domain, obtains the secondary newer estimate of reconstructed image；Again by two Secondary newer estimate is sent into routing selecting module 2, and routing selecting module 2 is square by secondary newer estimate feeding regularization Error minimum module 3, continues interative computation, achievees the purpose that gradually to promote output image quality.
2nd, method
1st, step is 1.：
The reconstructing method of the compressed sensing image is：
A, sparse coefficients of the compression sampling value y of the image of input under sparse dictionary D is solved η be adjustable regularisation parameter, sparse dictionary D can be discrete cosine transform, from Dissipate wavelet transformation or the dictionary based on study；
B, by above-mentioned sparse coefficientIt is multiplied with sparse dictionary D, the initial estimate z for obtaining image reconstruction is：
2nd, step is 3.：
A, the image overlap partition unit (401) of image filtering processing module (4) receives regularization mean square error minimum modulus The image x of block (3) output carries out overlap partition to x, obtains multiple image block x there are pixel overlappingi, it is similar to be sent into image Block group generation unit (402)；Image block xi=Ri(x), i=1,2 ... n, n are the integer more than 2, represent the total of image block Number, Ri() represents the operation matrix of the i-th image block of extraction, and tile size is generally chosen for B=8 rows × 8 and arranges；
B, each image block x that image similar block group generation unit (402) is sent into image overlap partition unit (401)i, Where it in defined neighborhood window of image x, ask for the most like image blocks of its m-1, m be 40~60 between it is just whole Number, the size of neighborhood window is 40 × 40；After the image block and its similar image block vector quantization, generation is corresponding with the image block Data matrix Ai, it is sent into transform domain and removes average soft-threshold de-noising unit (403)；Data matrixRepresent the i-th image block of extraction and its matrix of similar block composition, To extract the operation matrix of the similar block group of the i-th image block；
C, transform domain goes average soft-threshold de-noising unit (403) to be sent into image similar block group generation unit (402) every One data matrix Ai, dictionary Ψ is converted by its covariance matrix generation principal component analysis (PCA) first；Then dictionary will be converted Ψ and data matrix AiMultiplication obtains AiCoefficient in transform domain αi；Next to αiIt is corresponded to according to subband and calculates average ζiAnd variances sigmai, meter Calculate threshold valueFurther to factor alphaiAccording toCarry out threshold value shrink process；Finally to threshold value The coefficient obtained after shrink processInverse transformation is carried out, obtains the data matrix after filtering processIt is sent into figure As block pixel domain is averaging unit (404)；Here the point division operation of/expression Element-Level .* represent the point multiplication operation of Element-Level, Sgn () represents the symbol extraction computing of Element-Level, and max () expressions are maximized computing, and abs () represents that Element-Level takes definitely It is worth computing, ε is the small constant that is not zero for avoiding calculation overflow, Ψ-1Represent the inverse matrix of Ψ, N is the dimension of original image；
D, image block pixel domain is averaging unit (404) and goes what average soft-threshold de-noising unit (403) was sent into transform domain All data matrixesI=1,2 ..., n pass through calculatingObtain the two of reconstructed image Secondary newer estimate z is sent into routing selecting module (2)；HereForTransposition, 1B×mIt is B × m for scale, element It is worth the matrix for 1, B × m is matrixDimension.
3rd, emulation experiment
In order to prove the validity of the embodiment of the present invention, applicant has carried out emulation experiment under MATLAB platforms.Fig. 2 gives The present invention is gone out with document [2], [3] method compared with sample rate is arranged to 0.1 reconstructed image subjective vision effect, test chart As being 256 × 256Boats images；Fig. 2 .1 are experiment original images, and Fig. 2 .2 are the reconstruct images using document [2] method Picture, Fig. 2 .3 are the reconstructed images using document [3] method, and Fig. 2 .4 are the reconstructed images using the method for the present invention.It is testing In, compressed sensing calculation matrix is using accidental projection matrix；The size for setting the image block of image overlap partition is 8 × 8, Block and block at intervals of 4；In the extraction of similar image block group, the scale of structure similar image block group is set as m=60, search The window size of similar block is 40 × 40, sets regularisation parameter λ=0.09, μ=0.0025.

Claims (5)

1. a kind of compression of images sensing reconstructing system using group normalization rarefaction representation, it is characterised in that：
Including the processing of initialization module (1), routing selecting module (2), regularization mean square error minimum module (3) and image filtering Module (4)；
Initialization module (1), routing selecting module (2), regularization mean square error minimum module (3) and image filtering processing module (4) interact successively, image filtering processing module (4) and routing selecting module (2) interaction.
2. by compression of images sensing reconstructing system described in claim 1, it is characterised in that：
The image filtering processing module (4) includes image overlap partition unit (401) interactive successively, image similar block group Generation unit (402), transform domain go average soft-threshold de-noising unit (403) and image block pixel domain to be averaging unit (404).
3. the compression of images sensing reconstructing method based on claim 1,2 systems, it is characterised in that comprise the following steps：
1. initialization module (1) receives the compression sampling value y of input picture, and using the reconstructing method of compressed sensing image, it is raw Into the initial estimate z of compressed sensing image reconstruction, routing selecting module (2) is sent into, the graceful distance a=0 of initial setting up Donald Bragg, Iterations t=0, maximum iteration J, regularisation parameter λ and μ；
2. the compressed sensing image reconstruction that regularization mean square error minimum module (3) is inputted according to routing selecting module (2) is estimated The compression sampling value y of evaluation z and input picture, pass through solutionObtain reconstruct image The once newer estimate x of picture is sent into image filtering processing module (4)；Φ is calculation matrix, is usually chosen for random square Battle array, a are the graceful distance of Donald Bragg；
3. the once update for the reconstructed image that image filtering processing module (4) inputs regularization mean square error minimum module (3) Estimate x be filtered, obtain the secondary newer estimate z of reconstructed image；
4. update the graceful distance a of Donald Bragg：A=a- (x-z), update iterations t=t+1；If iterations t is no more than setting Maximum iteration J, then back to step 2.；Otherwise step is performed 5.；
5. terminating computing, the reconstructed image z finally obtained is exported.
4. the compression of images sensing reconstructing method as described in claim 3, it is characterised in that the step is 1.：
The reconstructing method of the compressed sensing image is：
A, sparse coefficients of the compression sampling value y of the image of input under sparse dictionary D is solved η be adjustable regularisation parameter, sparse dictionary D can be discrete cosine transform, from Dissipate wavelet transformation or the dictionary based on study；
B, by above-mentioned sparse coefficientIt is multiplied with sparse dictionary D, the initial estimate z for obtaining image reconstruction is：
5. the compression of images sensing reconstructing method as described in claim 3, it is characterised in that the step is 3.：
A, the image overlap partition unit (401) of image filtering processing module (4) receives regularization mean square error minimum module (3) The image x of output carries out overlap partition to x, obtains multiple image block x there are pixel overlappingi, it is sent into the life of image similar block group Into unit (402)；Image block xi=Ri(x), i=1,2 ... n, n are the integer more than 2, represent the sum of image block, Ri () represents the operation matrix of the i-th image block of extraction, and tile size is generally chosen for B=8 rows × 8 and arranges；
B, each image block x that image similar block group generation unit (402) is sent into image overlap partition unit (401)i, at it In the defined neighborhood window of place image x, its m-1 most like image blocks are asked for, m is the positive integer between 40~60, adjacent The size of domain window is 40 × 40；After the image block and its similar image block vector quantization, data corresponding with the image block are generated Matrix Ai, it is sent into transform domain and removes average soft-threshold de-noising unit (403)；Data matrixRepresent the i-th image block of extraction and its matrix of similar block composition, To extract the operation matrix of the similar block group of the i-th image block；
C, transform domain goes each that average soft-threshold de-noising unit (403) is sent into image similar block group generation unit (402) Data matrix Ai, dictionary Ψ is converted by its covariance matrix generation principal component analysis (PCA) first；Then will conversion dictionary Ψ with Data matrix AiMultiplication obtains AiCoefficient in transform domain αi；Next to αiIt is corresponded to according to subband and calculates average ζiAnd variances sigmai, calculate threshold valueFurther to factor alphaiAccording to Carry out threshold value shrink process；The finally coefficient to being obtained after threshold value shrink processInverse transformation is carried out, after obtaining filtering process Data matrixIt is sent into image block pixel domain and is averaging unit (404)；Here the point of/expression Element-Level is except fortune It calculates .* represents the point multiplication operation of Element-Level, and sgn () represents the symbol extraction computing of Element-Level, and max () expressions are maximized fortune It calculates, abs () represents that Element-Level takes absolute value computing, and ε is the small constant that is not zero for avoiding calculation overflow, Ψ-1Represent that Ψ's is inverse Matrix, N are the dimension of original image；
D, image block pixel domain averaging unit (404) removes transform domain average soft-threshold de-noising unit (403) feeding to own Data matrixI=1,2 ..., n pass through calculatingObtain reconstructed image it is secondary more New estimate z is sent into routing selecting module (2)；HereForTransposition, 1B×mFor scale be B × m, element value is 1 matrix, B × m are matrixDimension.
CN201711338419.8A 2017-12-14 2017-12-14 image compressed sensing reconstruction system and method using group normalization sparse representation Expired - Fee Related CN108093266B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201711338419.8A CN108093266B (en) 2017-12-14 2017-12-14 image compressed sensing reconstruction system and method using group normalization sparse representation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201711338419.8A CN108093266B (en) 2017-12-14 2017-12-14 image compressed sensing reconstruction system and method using group normalization sparse representation

Publications (2)

Publication Number Publication Date
CN108093266A true CN108093266A (en) 2018-05-29
CN108093266B CN108093266B (en) 2019-12-06

Family

ID=62175613

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201711338419.8A Expired - Fee Related CN108093266B (en) 2017-12-14 2017-12-14 image compressed sensing reconstruction system and method using group normalization sparse representation

Country Status (1)

Country Link
CN (1) CN108093266B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109345453A (en) * 2018-09-12 2019-02-15 中南民族大学 Utilize the image super-resolution reconfiguration system and method for standardization group Sparse rules
CN111193925A (en) * 2019-12-25 2020-05-22 杭州中威电子股份有限公司 Image compressed sensing coding and normalization method based on block vector inner product

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104159112A (en) * 2014-08-08 2014-11-19 哈尔滨工业大学深圳研究生院 Compressed sensing video transmission method and system based on dual sparse model decoding
CN105306936A (en) * 2015-07-17 2016-02-03 福州大学 BCS (Block Compressive Sensing)-based image coding method
EP3154022A1 (en) * 2015-10-07 2017-04-12 Univerza v Ljubljani A method of compressive sensing-based image filtering and reconstruction, and a device for carrying out said method
CN106651974A (en) * 2016-11-03 2017-05-10 中南民族大学 Image compressive sensing reconstruction system and method utilizing weighted structural group sparse regulation

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104159112A (en) * 2014-08-08 2014-11-19 哈尔滨工业大学深圳研究生院 Compressed sensing video transmission method and system based on dual sparse model decoding
CN105306936A (en) * 2015-07-17 2016-02-03 福州大学 BCS (Block Compressive Sensing)-based image coding method
EP3154022A1 (en) * 2015-10-07 2017-04-12 Univerza v Ljubljani A method of compressive sensing-based image filtering and reconstruction, and a device for carrying out said method
CN106651974A (en) * 2016-11-03 2017-05-10 中南民族大学 Image compressive sensing reconstruction system and method utilizing weighted structural group sparse regulation

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109345453A (en) * 2018-09-12 2019-02-15 中南民族大学 Utilize the image super-resolution reconfiguration system and method for standardization group Sparse rules
CN109345453B (en) * 2018-09-12 2022-12-27 中南民族大学 Image super-resolution reconstruction system and method utilizing standardization group sparse regularization
CN111193925A (en) * 2019-12-25 2020-05-22 杭州中威电子股份有限公司 Image compressed sensing coding and normalization method based on block vector inner product

Also Published As

Publication number Publication date
CN108093266B (en) 2019-12-06

Similar Documents

Publication Publication Date Title
Ng et al. Solving constrained total-variation image restoration and reconstruction problems via alternating direction methods
Sankaranarayanan et al. Compressive acquisition of dynamic scenes
CN106651974B (en) Utilize the compression of images sensing reconstructing system and method for weighting structures group Sparse rules
Li et al. Universal HMT based super resolution for remote sensing images
CN110675347B (en) Image blind restoration method based on group sparse representation
CN106952317B (en) Hyperspectral image reconstruction method based on structure sparsity
Hirakawa et al. Skellam shrinkage: Wavelet-based intensity estimation for inhomogeneous Poisson data
Zhao et al. Image compressive-sensing recovery using structured laplacian sparsity in DCT domain and multi-hypothesis prediction
CN111340697A (en) Clustering regression-based image super-resolution method
CN108093266A (en) Utilize a group compression of images sensing reconstructing system and method for normalization rarefaction representation
Ongie et al. A fast algorithm for structured low-rank matrix recovery with applications to undersampled MRI reconstruction
Alvarez-Ramos et al. Image super-resolution via two coupled dictionaries and sparse representation
CN107133921B (en) Image super-resolution reconstruction method and system based on multilevel neighborhood embedding
Taşmaz et al. Image enhancement via space-adaptive lifting scheme exploiting subband dependency
Dogiwal et al. Super resolution image reconstruction using wavelet lifting schemes and gabor filters
CN111932473B (en) Multi-resolution sparse coding phase information noise reduction algorithm and storage medium
Suryanarayana et al. Single image super-resolution algorithm possessing edge and contrast preservation
Gou et al. Image super‐resolution based on the pairwise dictionary selected learning and improved bilateral regularisation
Abrardo et al. A compressive sampling scheme for iterative hyperspectral image reconstruction
US20060245667A1 (en) 1-D image restoration using a sliding window method
Hirakawa et al. “Rewiring” Filterbanks for Local Fourier Analysis: Theory and Practice
Hattay et al. Geometric blind source separation using adaptive lifting scheme
Gomeztagle et al. Super-resolution method based on wavelet atomic functions in images and video sequences
Ebrahimi et al. Self-similarity in imaging 20 years after “Fractals everywhere”
CN117312837B (en) Minimum total variation-oriented signal reconstruction method and system

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
CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20191206

Termination date: 20211214