CN105654530B - A kind of high robust image adaptive compression method based on compressed sensing - Google Patents
A kind of high robust image adaptive compression method based on compressed sensing Download PDFInfo
- Publication number
- CN105654530B CN105654530B CN201610129938.2A CN201610129938A CN105654530B CN 105654530 B CN105654530 B CN 105654530B CN 201610129938 A CN201610129938 A CN 201610129938A CN 105654530 B CN105654530 B CN 105654530B
- Authority
- CN
- China
- Prior art keywords
- image
- weight
- image block
- measured value
- fritter
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/003—Reconstruction from projections, e.g. tomography
- G06T11/006—Inverse problem, transformation from projection-space into object-space, e.g. transform methods, back-projection, algebraic methods
-
- 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
Abstract
A kind of high robust image adaptive compression method based on compressed sensing, is related to image procossing.Original image pre-processes；Initialization survey battle array and weight factor；It calculates the significance of each image block and updates weight factor；Judge whether each fritter contains target, and updates weight factor；It is designed according to significant factor and measures battle array；Image measurement and transmission；Measured value reparation；Image reconstruction.Using the total variance norm suitable for the not sparse image of time domain, it is capable of detecting when feature and inapparent target, is suitble to compress image；Accomplish that spatial domain directly measures, first image need not be carried out the operation such as converting, the pendulous frequency distribution of use is simple according to algorithm, avoid the operation of complicated and time consumption, can threshold value be set to the amplitude size of higher-order spectrum, judge the presence or absence of target in image, is suitble to the higher underwater sonar Image Communication scene of requirement of real-time.Contribute in the case where not reducing reconstructed image quality, reduces compression ratio, and improve robustness.
Description
Technical field
The present invention relates to image procossings, are compressed more particularly, to a kind of high robust image adaptive based on compressed sensing
Method.
Background technology
Since there is image voice and text to be difficult to the visual effect stated, in recent years the multimedia services such as image gradually at
For the hot spot of Communication Studies, exploitation optimization digital picture and Digital Video System are one extremely important under severe channel conditions
With the work rich in challenge, image transmission data amount is huge, and picture quality is more using the subjective feeling of people as standard, therefore one
Kind can improve people subjective feeling under the premise of compression of images mode have very strong practical significance.
2006, the base of Dohono, Candes and scientist Tao of Chinese origin et al. in Signal approximation and rarefaction representation scheduling theory
It is theoretical that a kind of i.e. compressed sensing CS of completely new signal acquisition theory (Compressed Sensing) is established on plinth.If some
The only a small amount of nonzero element of vector just claims this vector to have sparsity.Compressive sensing theory is pointed out, as long as signal is at some
Specific transform domain is sparse, then can be projected to high dimensional signal with one and the transformation incoherent observing matrix of base
On one lower dimensional space, then utilize the priori of sparsity by solving an optimization problem from known a small amount of projection value
In original signal reconstructed with high probability.For the channel of inclement condition, pass through less measured value weight since compressed sensing has
The feature of structure signal, and measured value does not have the difference of importance, therefore compressed sensing realizes image under severe channel conditions
High robust compression transmission have some superiority.
Traditional compression method based on compressed sensing does not consider the significant properties of image, to different significant properties
Image block uses the same measured rate, and such method cannot be played carries out special protection to the interested marking area of human eye
Effect.It is adaptive according to the significance degree if the significance degree of image different zones can be distinguished according to the different level of interest of human eye
The measured rate of different masses should be distributed, each piece of level of data compression is dynamically adjusted, after compression ratio and compression can be effectively improved
The visual effect of image.
Severe channel condition can cause the measured value of transmission to damage, and impaired measured value can shadow when being reconstructed
Optimum results are rung, image is caused a degree of deterioration occur.Due to having certain relevance between measured value, and develop into
Ripe Error Correction of Coding allows the location information of error code to obtain, therefore can by the relevance between error code position and measured value
To realize the robustness enhancing of compressed encoding.
Invention content
The purpose of the present invention is for the existing above problem of image transmitting design under severe channel conditions, provide in reality
While existing compression of images, ensure the reservation of the reservation of important information, especially conspicuousness compared with weak signal target information as far as possible, and add
A kind of high robust image adaptive compression method based on compressed sensing of the robustness of strong compression method.
For convenience of description present disclosure, the definition below in connection with term is provided first：
1 is defined, total variance norm is introduced, is rebuild using the compressed sensing that conjugate gradient method carries out：
For the signal f that time domain is sparse, f is the matrix of a M rows N row, compressed sensing reconstruction and optimization problem in the present invention
It can be expressed with following formula：
min||f||1s.t.M0F=y. (1)
In formula (1), M0To measure battle array, y is measured value.The estimated value of f is acquired to the method optimized by measured valueIt then needs to solve following formula：
The process of formula (2) is solved using conjugate gradient method can be written as formula (3)：
Wherein μiIt is iteration step length, fiIt represents obtained by ith iteration as a result, ▽ L (fi) be defined as follows：
Represent matrix M0Moore-Penrose is inverse or pseudoinverse.Each direction of search of conjugate gradient method is mutual
Conjugation, and these directions of search are only the combination in negative gradient direction and the direction of search of last iteration, overcome steepest
Descent method restrains slow disadvantage, in turn avoids the shortcomings that Newton method needs storage and calculates Hesse matrixes and invert, has required
Amount of storage is small, walks convergence, and stability is high and the advantages of not needing any external parameter.
If f does not have sparsity in time domain, solves following expression formula and optimal reconfiguration image may be implemented：
TV (f) is the total variance of image f, and expression formula is as follows：
Parameter lambda ∈ [0.001,0.01], horizontal gradient operatorAnd vertical gradient operatorSuch as formula (7) and (8) institute
Show：
Wherein N is the columns of image array.Specific steps are referring to document [1]:The Physics of Compressive
Sensing and the Gradient-Based Recovery Algorithms。
2 are defined, reconstructed image quality：
Reconstructed image quality is characterized with SSIM values, the calculation formula of SSIM is as follows：
Wherein：
Wherein x, y respectively represent the reconstructed image of original image and quality to be tested and assessed, xiOriginal image pixels value is represented,
yiQuality reconstruction image pixel value to be tested and assessed is represented, N is image total pixel number, and C1, C2 are custom parameter, according to image reality
Situation is selected, and when evaluating and testing picture quality in the present invention, selects C1=1.275, C2=3.825.Specific steps are referring to document
[2]:Objective video quality assessment。
3 are defined, compression ratio：
In the present invention, defining compression ratio isThe ratio between i.e. compressed data volume and original data volume.
The present invention includes the following steps：
Step 1, original image pre-processes：
It is that L × W original images are divided into N=(L × W)/m by resolution ratio2A non-overlapping copies, size are the image block of m*m,
Wherein m=2n, n >=3 are denoted as I1,I2,…,Ii,…IN, here L represent the line number i.e. length of original image, W represents original image
Columns, that is, width, N represent caused by image block number, m represent caused by each square image blocks the length of side, Ii
Represent i-th of image block, the index of i representative image blocks, i ∈ { 1,2,3 ..., N }；
Step 2, initialization survey battle array and weight factor：
Define initial measurement battle array UL×W, it is a L row, W is arranged, and element is all 1 matrix, initialization weight factor Weight
(i)=1, i is the index of image block, i ∈ { 1,2,3 ..., N }；
Step 3, it calculates the significance of each image block and updates weight factor：
The fast significance degree of image is calculated according to the spatial domain activity of image, and updates Weight (i), calculates each fritter
Spatial domain activity, be denoted as IAM0, according to spatial domain activity IAM0, the weight factor for distributing to each fritter is as follows：
0≤IAM03 Weight (i)=0.1 ＜
3≤IAM05 Weight (i)=0.3 ＜
5≤IAM07 Weight (i)=0.5 ＜
7≤IAM010 Weight (i)=0.6 ＜
10≤IAM012 Weight (i)=0.7 ＜
12≤IAM0Weight (i)=0.8
Step 4, judge whether each fritter contains target, and update weight factor：
The bispectrum peak value for calculating each fritter determines whether each fritter contains target according to bispectrum peak value, and method is as follows：
1) it is one group of observation data, representative image block I to enable x (1), x (2) ..., x (N')iThe pixel value of each pixel, N'
For image block IiTotal number of pixels, i.e. N'=m2, and set f0It is sample frequency, N0Total frequency sampling number, then Δ f=f0/
N0It is the frequency sampling interval in bispectrum region on both horizontally and vertically.
2) by be divided into K sections to data, every section contains M observation sample, i.e. N'=KM, and to every segment data y'(i)(n)
Subtract the mean value y of this section(i)(n)=y'(i)(n)-E[y'(i)(n)]；
3) DFT coefficient is calculatedK=0 ..., M/2；I=1 ..., K, wherein y(i)(n)
It is that the i-th segment data subtracts the value after this section of mean value；
4) triple correlation of DFT coefficient is calculated
Wherein p1、p2Offset when to calculate related, 0≤k2≤k1,k1+k2≤f0/2；
5) provided by the average value of K sections of bi-spectrum estimations to the bi-spectrum estimations of data, i.e.,
Wherein
6) Bispectrum characteristic parameter t (i)=max [B (f of the bispectrum peak value of each image block as the image block is read1,
f2)], i indexes for foregoing image block, and to t (i) according to sorting from big to small, sequence index is denoted as Idx (i), bispectrum peak
It is worth maximum four image blocks, the possibility containing target is maximum, update Weight (i)=1 | and Idx (i)≤4 }, residual image
The weight factor of block remains unchanged.
Step 5, it is designed according to significant factor and measures battle array：
By initial measurement battle array UL×WIt is divided into N=(L × W)/m2A non-overlapping copies, size are the fritter of m*m, are denoted as um×m (i),
Wherein m=2n, n >=3, i are measurement battle array fritter call number corresponding with image block index, i ∈ { 1,2,3 ..., N }.um×m (i)For m
Row, m row, element are all 1 matrix.Define a m row, the random matrix of m row
Element in Ψ, which meets, to be uniformly distributed, and idx represents the index of element in Ψ, according to weight factor, then measures each of gust fritter
ElementWherein symbolIndicate the downward rounding of each element to the matrix in symbol,
Idx represents um×m (i)The index of middle element, idx=1,2 ... m × m, by um×m (i)The measurement battle array of L × W is combined into according to index i
UL×W。
Step 6, image measurement and transmission：
With measurement battle array UL×WTraditional compressed sensing sampling is carried out to image, obtains measured value, and measured value is described with more
Coding carries out message sink coding and channel error correction coding, and is sent into transmission.
Step 7, measured value reparation：
According to the errors present combination message source and channel decoding of Error Correction of Coding prompt, the impaired location index of measured value is obtained,
Due to having certain relevance between measured value, interpolation reparation is carried out using the measured value on its periphery to being damaged measured value, is repaiied
Compound method is as follows, takes the small window of 3 × 3 centered on measured value in damaged condition, and interpolation is carried out to small window operator shown in table 1；
Table 1
| 1 | 1 | 1 |
| 1 | 1/8 | 1 |
| 1 | 1 | 1 |
Step 8, image reconstruction：
Total variance norm is introduced, using conjugate gradient method, image reconstruction is carried out according to the measurement value matrix after reparation.
In the present invention, weight factor is the compression ratio of each image block, and the reduced overall rate of image is
Present invention introduces image spatial domain activity and image bispectrum characteristics, from two spatial domain, frequency domain angle recognition image weights
Region is wanted, the region pendulous frequency is adaptively adjusted, important information is remained to the maximum extent while improving compression ratio.In conjunction with
Compressed encoding based on compressed sensing is encoded with channel error correction, alignment measurement damaged location, around impaired measured value
Measured value carries out interpolation, repairs and measures value matrix, improves the robustness of compressed encoding.Wherein, picture activity is utilized in spatial domain,
It is according to the significance degree for judging image-region with amplitude；Judge to scheme for foundation with amplitude and phase using bispectrum characteristic in frequency domain
Whether picture region contains target, according to the pendulous frequency in this two layers of information self-adapting distribution different images region.It is proposed by the present invention
Method uses the total variance norm suitable for the not sparse image of time domain, is capable of detecting when feature and inapparent target, is suitble to
Image is compressed；Accomplish that spatial domain directly measures, first image need not be carried out operation, the pendulous frequency point of use such as to convert
It is simple with foundation algorithm, the operation of complicated and time consumption is avoided, especially for sonar image, in sonar image if without mesh
Mark, image is large-scale continuum；If there is target exists, then it will appear small-scale continuum and target and the back of the body
Transition between scape, therefore the variation of amplitude and phase will be reflected into the higher-order spectrum of image, it can be to the width of higher-order spectrum
It spends size and threshold value is set, to judge the presence or absence of target in image, therefore the present invention is very suitable for the higher underwateracoustic of requirement of real-time
Receive Image Communication scene.
The present invention adjusts the image district by the significance degree of two layers of feature differentiation image-region according to significance degree dynamic
The pendulous frequency in domain retains the important information of image as much as possible while improving compression ratio；The present invention helps do not reducing
In the case of reconstructed image quality, compression ratio is reduced, and improve robustness, is suitable for the bad underwater sound communication field of channel condition
Scape.
Description of the drawings
Fig. 1 is the measurement system of battle formations picture in the embodiment of the present invention.
Specific implementation mode
Following embodiment will the present invention is further illustrated in conjunction with attached drawing.
The saturating sonar image of 3000 sound of ARIS Explorer for being 256 × 256 to resolution ratio carries out compressed sensing, according to figure
Image is divided into 8 × 8 fritter by the size of picture.The significance degree of image block is calculated according to the spatial domain activity of image, and is updated
Weight (i) calculates the spatial domain activity of each fritter, and according to spatial domain activity, the measured rate for distributing to each fritter is as follows：
0≤IAM03 Weight (i)=0.1 ＜
3≤IAM05 Weight (i)=0.3 ＜
5≤IAM07 Weight (i)=0.5 ＜
7≤IAM010 Weight (i)=0.6 ＜
10≤IAM012 Weight (i)=0.7 ＜
12≤IAM0Weight (i)=0.8
Calculate the bispectrum B (f of each fritter1,f2), read bispectrum of the bispectrum peak value of each image block as the image block
Characteristic parameter t (i)=max [B (f1,f2)], i be foregoing image block index, to t (i) according to bispectrum peak value from big to small
Sequence, sequence index are denoted as Idx (i), and maximum four image blocks of bispectrum peak value, the possibility containing target is maximum, enables
Weight (i)=1 | Idx (i)≤4 }, the weight factor of residual image block remains unchanged.
Initial measurement battle array is divided into N=(L × W)/m2A non-overlapping copies, size are the fritter of m × m, are denoted as um×m (i),
Middle m=2n, n >=3, i are image block call number corresponding with image index.um×m (i)For m rows, m row, element is all 1 matrix.
Define a m row, the random matrix of m rowElement in Ψ meets uniformly
Distribution, idx represents the index of element in Ψ, according to weight factor, then measures each element of gust fritterWherein symbolIndicate the downward rounding of each element to the matrix in symbol, idx
Represent um×m (i)The index of middle element, idx=1,2 ... m × m, by um×m (i)The measurement battle array U of L × W is combined into according to index iL×W。
UL×WAs shown in Figure 1.
As shown in Figure 1, the point of black is the part not sampled, the point of black is more intensive to illustrate that pendulous frequency is lower, according to
Battle array is measured, after adaptive measuring, the compression ratio of image data is 0.38, reconstructed image quality 0.9569, and without adaptive
It should measure, to reach identical reconstructed image quality, the compression ratio needed is 0.6, illustrates that the present invention helps to improve performance, subtracts
Few transmitted data amount.
With measurement battle array UL×WTraditional compressed sensing sampling is carried out to image, obtains measured value, and measured value is described with more
Coding carries out message sink coding and channel error correction coding, and is sent into transmission.
According to the errors present combination message source and channel decoding of Error Correction of Coding prompt, the impaired location index of measured value is obtained,
Due to having certain relevance between measured value, interpolation reparation is carried out using the measured value on its periphery to being damaged measured value, is repaiied
Compound method is as follows, takes the small window of 3 × 3 centered on measured value in damaged condition, to small window measured value repair operator shown in table 1.
Under underwater acoustic channel, the unconditional packet loss of GB Network Packet Loss models is ulp=0.0176, condition packet loss clp=
0.2565, reconstructed image quality is 0.7217 after measured value packet loss, and it is 0.9499 to measure reconstructed image quality after value interpolation is repaired,
Compression reconfiguration picture quality when non-packet loss is 0.9569, and the picture quality after interpolation reparation is compared before relatively repairing, and is increased.
Claims (1)
1. a kind of high robust image adaptive compression method based on compressed sensing, it is characterised in that include the following steps：
Step 1, original image pre-processes：
It is that L × W original images are divided into N=(L × W)/m by resolution ratio2A non-overlapping copies, size are the image block of m*m, wherein m
=2n, n >=3 are denoted as I1,I2,…,Ii,…IN, here L represent the line number i.e. length of original image, W represents the row of original image
It is width to count, and N represents the number of generated image block, and m represents the length of side of generated each square image blocks, IiIt represents
I-th of image block, the index of i representative image blocks, i ∈ { 1,2,3 ..., N }；
Step 2, initialization survey battle array and weight factor：
Define initial measurement battle array UL×W, it is a L row, W row, element is all 1 matrix, initialization weight factor Weight (i)=
1, i is the index of image block, i ∈ { 1,2,3 ..., N }；
Step 3, it calculates the significance of each image block and updates weight factor：
The significance degree of image block is calculated according to the spatial domain activity of image, and updates Weight (i), calculates the sky of each fritter
Domain activity, is denoted as IAM0, according to spatial domain activity IAM0, the weight factor for distributing to each fritter is as follows：
0≤IAM03 Weight (i)=0.1 ＜
3≤IAM05 Weight (i)=0.3 ＜
5≤IAM07 Weight (i)=0.5 ＜
7≤IAM010 Weight (i)=0.6 ＜
10≤IAM012 Weight (i)=0.7 ＜
12≤IAM0Weight (i)=0.8
Step 4, judge whether each fritter contains target, and update weight factor：
The bispectrum peak value for calculating each fritter determines whether each fritter contains target according to bispectrum peak value, and method is as follows：
1) it is one group of observation data, representative image block I to enable x (1), x (2) ..., x (N')iThe pixel value of each pixel, N' is image
Block IiTotal number of pixels, i.e. N'=m2, and set f0It is sample frequency, N0Total frequency sampling number, then Δ f=f0/N0Be
Frequency sampling interval of the bispectrum region on both horizontally and vertically；
2) by be divided into K sections to data, every section contains M observation sample, i.e. N'=KM, and to every segment data y'(i)(n) it subtracts
The mean value y of this section(i)(n)=y'(i)(n)-E[y'(i)(n)]；
3) DFT coefficient is calculatedK=0 ..., M/2；I=1 ..., K, wherein y(i)(n) it is
I segment datas subtract the value after this section of mean value；
4) triple correlation of DFT coefficient is calculatedIts
Middle p1、p2Offset when to calculate related, 0≤k2≤k1,k1+k2≤f0/2；
5) provided by the average value of K sections of bi-spectrum estimations to the bi-spectrum estimations of data, i.e.,Wherein
6) Bispectrum characteristic parameter t (i)=max [B (f of the bispectrum peak value of each image block as the image block is read1,f2)], i
It is indexed for foregoing image block, to t (i) according to sorting from big to small, sequence index is denoted as Idx (i), and bispectrum peak value is maximum
Four image blocks, possibility containing target is maximum, update Weight (i)={ 1 | Idx (i)≤4 }, the power of residual image block
Repeated factor remains unchanged；
Step 5, it is designed according to significant factor and measures battle array：
By initial measurement battle array UL×WIt is divided into N=(L × W)/m2A non-overlapping copies, size are the fritter of m*m, are denoted as um×m (i), wherein m
=2n, n >=3, i are measurement battle array fritter call number corresponding with image block index, i ∈ { 1,2,3 ..., N }, um×m (i)For m rows, m
Row, element are all 1 matrix, define a m row, the random matrix Ψ of m row:
Element in Ψ, which meets, to be uniformly distributed, and idx represents the index of element in Ψ, according to weight factor, then measures each of gust fritter
ElementWherein symbolIndicate the downward rounding of each element to the matrix in symbol,
Idx represents um×m (i)The index of middle element, idx=1,2 ... m × m, by um×m (i)The measurement battle array of L × W is combined into according to index i
UL×W；
Step 6, image measurement and transmission：
With measurement battle array UL×WTraditional compressed sensing sampling is carried out to image, obtains measured value, and by measured value with multiple description coded
Message sink coding and channel error correction coding are carried out, and is sent into transmission；
Step 7, measured value reparation：
According to the errors present combination message source and channel decoding of Error Correction of Coding prompt, the impaired location index of measured value is obtained, due to
There is certain relevance between measured value, interpolation reparation, reparation side are carried out using the measured value on its periphery to being damaged measured value
Method is as follows, takes the small window of 3 × 3 centered on measured value in damaged condition, and interpolation is carried out to small window operator shown in table 1；
Table 1
Step 8, image reconstruction：
Total variance norm is introduced, using conjugate gradient method, image reconstruction is carried out according to the measurement value matrix after reparation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610129938.2A CN105654530B (en) | 2016-03-08 | 2016-03-08 | A kind of high robust image adaptive compression method based on compressed sensing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610129938.2A CN105654530B (en) | 2016-03-08 | 2016-03-08 | A kind of high robust image adaptive compression method based on compressed sensing |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105654530A CN105654530A (en) | 2016-06-08 |
| CN105654530B true CN105654530B (en) | 2018-07-31 |
Family
ID=56492341
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610129938.2A Active CN105654530B (en) | 2016-03-08 | 2016-03-08 | A kind of high robust image adaptive compression method based on compressed sensing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105654530B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108259909B (en) * | 2018-02-09 | 2020-09-01 | 福州大学 | Image compression method based on saliency object detection model |
| CN108335335B (en) * | 2018-02-11 | 2019-06-21 | 北京大学深圳研究生院 | A kind of point cloud genera compression method based on enhancing figure transformation |
| CN108548957B (en) * | 2018-05-23 | 2020-08-07 | 西北工业大学 | Dual-spectrum analysis method based on combination of cyclic modulation spectrum and piecewise cross correlation |
| CN110149516A (en) * | 2019-05-29 | 2019-08-20 | 东北石油大学 | A kind of compression sensing method of image, apparatus and system |
| CN112508089B (en) * | 2020-12-03 | 2023-10-31 | 国网山西省电力公司晋城供电公司 | Self-adaptive compressed sensing method for partial discharge signal compressed transmission |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102568016A (en) * | 2012-01-03 | 2012-07-11 | 西安电子科技大学 | Compressive sensing image target reconstruction method based on visual attention |
| CN105225207A (en) * | 2015-09-01 | 2016-01-06 | 中国科学院计算技术研究所 | A kind of compressed sensing imaging based on observing matrix and image rebuilding method |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150098639A1 (en) * | 2013-10-08 | 2015-04-09 | pita4 mobile LLC | Image Processing System, Image Processing Method and Image Processing Program |
| US20150116563A1 (en) * | 2013-10-29 | 2015-04-30 | Inview Technology Corporation | Adaptive Sensing of a Programmable Modulator System |
-
2016
- 2016-03-08 CN CN201610129938.2A patent/CN105654530B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102568016A (en) * | 2012-01-03 | 2012-07-11 | 西安电子科技大学 | Compressive sensing image target reconstruction method based on visual attention |
| CN105225207A (en) * | 2015-09-01 | 2016-01-06 | 中国科学院计算技术研究所 | A kind of compressed sensing imaging based on observing matrix and image rebuilding method |
Non-Patent Citations (3)
| Title |
|---|
| A Novel Vision-Based Adaptive Scanning for the Compression of Remote Sensing Images;C Shi 等;《IEEE Transactions on Geoscience and Remote Sensing》;20151006;第54卷(第3期);1366-1348 * |
| 分块压缩感知方法研究;许剑波;《万方数据》;20151229;1-54 * |
| 基于显著性信息的图像块压缩感知编码方法研究;胡开云;《中国优秀硕士学位论文全文数据库 信息科技辑》;20140515;I136-192 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105654530A (en) | 2016-06-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105654530B (en) | A kind of high robust image adaptive compression method based on compressed sensing | |
| WO2019213986A1 (en) | Multi-angle adaptive intra-frame prediction-based point cloud attribute compression method | |
| US11122293B2 (en) | Intra-frame prediction-based point cloud attribute compression method | |
| US11126887B2 (en) | Enhanced graph transformation-based point cloud attribute compression method | |
| CN108259916B (en) | Best match interpolation reconstruction method in frame in a kind of distributed video compressed sensing | |
| CN107040771B (en) | A kind of Encoding Optimization for panoramic video | |
| CN104376565A (en) | Non-reference image quality evaluation method based on discrete cosine transform and sparse representation | |
| CN108419083B (en) | Image multilevel wavelet full subband compressed sensing coding method | |
| CN109309840A (en) | Compressed sensing image encoding method based on noise shaping, storage medium | |
| CN106941609A (en) | The video-frequency compression method perceived based on self adaptation splits' positions | |
| CN109819260A (en) | Video steganography method and device based on the fusion of multi-embedding domain | |
| CN102946539B (en) | Method for estimating motion among video image frames based on compressive sensing | |
| Chen et al. | Adaptive underwater image compression with high robust based on compressed sensing | |
| CN108573509B (en) | Self-adaptive compressed sensing image coding method based on measurement domain significance detection model | |
| Zhang et al. | Fractal color image compression using vector distortion measure | |
| CN107707917B (en) | A kind of video adaptive sample rate setting method based on comentropy | |
| CN103186891A (en) | Hexagon image reconstruction method based on compressed sensing | |
| Yuan-Yuan et al. | Discrete cosine transform optimization in image compression based on genetic algorithm | |
| CN106530365B (en) | Self-adapting compressing sensing reconstructing method based on image information content difference | |
| CN108495134A (en) | Bayer image compression method based on JPEG2000 standards | |
| CN115294010A (en) | Method for evaluating quality of reference point cloud based on support vector machine | |
| CN105160664B (en) | Compressed sensing video reconstruction method based on low-rank model | |
| CN107592537B (en) | A kind of self-adapting compressing sampling distribution method towards Aerial Images collection | |
| Wu et al. | DCSN-Cast: deep compressed sensing network for wireless video multicast | |
| CN105704497A (en) | Fast select algorithm for coding unit size facing 3D-HEVC |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |