CN104199627B - Gradable video encoding system based on multiple dimensioned online dictionary learning - Google Patents
Gradable video encoding system based on multiple dimensioned online dictionary learning Download PDFInfo
- Publication number
- CN104199627B CN104199627B CN201410331199.6A CN201410331199A CN104199627B CN 104199627 B CN104199627 B CN 104199627B CN 201410331199 A CN201410331199 A CN 201410331199A CN 104199627 B CN104199627 B CN 104199627B
- Authority
- CN
- China
- Prior art keywords
- multiple dimensioned
- training set
- dictionary
- video
- sub
- 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
Abstract
The invention provides a kind of gradable video encoding system based on multiple dimensioned online dictionary learning, wherein：Stratification sparsity structure on wavelet transformation acquisition image different scale is utilized based on the sparse multiple dimensioned training set constructing module of stratification, elementary area in oriented energy acquisition image is extracted by Difference of Gaussian filter group, the image block for intercepting elementary area generates multiple dimensioned training set, online dictionary learning module ensure that the iteration optimization dictionary atom under low complex degree using stochastic gradient descent method, generate sub- dictionary base corresponding to multiple dimensioned training set, across yardstick frame of video reconstructed module arrives the loss high-frequency information of different levels to low-frequency video frame by the sub- dictionary base study of construction, reconstructed by the wavelet inverse transformation of different series, realize the gradable purpose of video quality.Present invention reduces the complexity of the super-resolution algorithms based on study, compared to reconstruction quality gain is H.264 achieved under different transmission rates, possesses good scalability.
Description
Technical field
The present invention relates to gradable video encoding scheme, in particular it relates to a kind of based on multiple dimensioned online dictionary learning
Gradable video encoding system.
Background technology
Perfect with HEVC standard, the formulation of HEVC graduated encoding scheme has been similarly subjected to extensive concern.For
The adaptive transmission of video requirement that meets on the heterogeneous network of different transmission properties and the application requirement of different clients, depending on
The gradability of frequency coding has very high theoretical research and actual application value.From H.264/AVC to HEVC, increasingly into
Ripe interframe and intra-frame prediction method improves rate distortion performance, such as adaptive kernel function：MDDT, ROT and adaptive
DCT/DST conversion etc., its essence be how effectively by analyze dictionary base or learn dictionary base linear combination it is effective and dilute
The problem of expressing natural sign thinly, most may be used in decoding end by adjacent or present frame local or non-local sample block to predict
The video image blocks of energy.Super-resolution reconstruction technology popular in recent years demonstrates dictionary learning algorithm can be effectively by non-
The mode of parameter is estimated the correlation between the low-resolution image and high-definition picture of sparse sampling.
Find that the gradable video encoding scheme of existing two main flows is Asia by the literature search to prior art
Microsoft Research Ruiqin Xiong, Jizheng Xu, Feng Wu and Shipeng Li were in 2007《IEEE
Transactions on Circuits and Systems for Video Technology》(TCSVT) delivered on periodical
" propose in the texts of Barbell-Lifting Based3-D Wavelet Coding Scheme " one and become based on 3-D wavelet
The gradable video encoding framework of classification, and German Heinrich-Hertz Institute (HHI) Schwarz are changed,
H.Marpe, D.Wiegand, T. were in 2007 years《IEEE Transactions on Circuits and Systems for
Video Technology》(TCSVT) " the Overview of the Scalable Video Coding delivered on periodical
Extension of the H.264/AVC propose in the texts of Standard " one based on encoding scheme H.264.It is but traditional
Dct transform and 2-D wavelet transforms can only capture the one-dimensional singular point at edge, can not effectively represent two dimension or even
The slickness on dimensional images border.Therefore Elad et al. in 2011《IEEE Journal of Selected Topics
in Signal Processing》On " the Multi-Scale Dictionary Learning Using Wavelets " that deliver
The multiple dimensioned study dictionary method based on wavelet decomposition is proposed in one text, is layered sparse structure by constructing, adaptively
Different scale images information is learnt, the fixed base of binding analysis dictionary and the adaptive base for learning dictionary, natural
Realize to natural signals yardstick asymptotics, at the same obtain than single scale learn dictionary base and the more preferable Its Sparse Decomposition of wavelet basis and
Rebuild effect.But the multiple dimensioned study dictionary is only applicable to the rarefaction representation of image, can not apply in Video coding and compression.
The content of the invention
For in the prior art the defects of, it is an object of the invention to provide it is a kind of based on multiple dimensioned online dictionary learning can
Scalable video coding system, vision signal code efficiency and the subjective and objective quality of reconfiguration system can be effectively improved, and can made
For a kind of general quality scalability video coding tool.
To realize object above, the present invention provides a kind of gradable video encoding system based on multiple dimensioned online dictionary learning
System, including：Based on the sparse multiple dimensioned training set constructing module of stratification, online dictionary learning module and across yardstick frame of video
Reconstructed module, wherein：
The multiple dimensioned training set constructing module sparse based on stratification obtains image difference using multi-level wavelet transform
Stratification sparsity structure on yardstick, the elementary area in oriented energy acquisition image is extracted by Difference of Gaussian filter group,
The image block for intercepting elementary area generates multiple dimensioned training set；
The online dictionary learning module ensure that the iteration optimization dictionary under low complex degree using stochastic gradient descent method
Atom, dictionary learning, multiple dimensioned sub- word corresponding to generation are carried out to the training set of different scale by online dictionary learning algorithm
Allusion quotation base；
Across the yardstick frame of video reconstructed module arrives different levels to low-frequency video frame by the sub- dictionary base study of construction
Loss high-frequency information, reconstructed by the wavelet inverse transformations of different series, realize the gradable purpose of video quality.
Preferably, the multiple dimensioned training set constructing module sparse based on stratification, the module are realized by reconstructing
Key frame on K-1 rank high-frequency sub-bands on the low frequency sub-band and 3 directions of picture are obtained by K rank wavelet transformations, for every
One sub- band logical crosses the extraction of Gaussian filter progress primitive blocks and direction is classified, the primitive in each direction class of different scale
Block corresponds to a sub- training set, while has stratification sparsity structure on the dictionary base for training on sub- training set to obtain.
It is highly preferred that the multiple dimensioned training set construction based on wavelet transformation can pass through wavelet transformation, primitive blocks
Extraction and sorting technique are operable to realize to whole reconstructed key-frame.
Preferably, described online dictionary learning module, the module is realized realizes rarefaction representation by stochastic gradient descent method
The minimum of error, the immanent structure for representing high dimensional signal of its property of can adapt to can be more effectively dilute relative to fixed base
Dredge and represent vision signal, this rarefaction representation on excessively complete study dictionary basic matrix has structural sparse.
It is highly preferred that described online dictionary learning module can be based only upon current training block minimum in each iteration
Change cost function, reduce computation complexity and space utilization rate, act solely on different sub- training set groups obtain it is different
Sub- dictionary pair.
Preferably, across the yardstick frame of video reconstructed module is found by a kind of convex relaxed algorithm model realization
The sparse sub- dictionary base for being multiplied by corresponding dictionary learning and obtaining of optimal rarefaction representation, it is exactly by the wavelet inverse transformation of different rank
Obtain gradable reconstruction signal.
Present system provides general solution for the coding compression of vision signal.It is used in the present invention to be based on
The sparse multiple dimensioned training set constructing module of stratification can be by wavelet transformation, primitive blocks extraction and sorting technique to whole weight
Structure key frame is operated what is obtained, and it is sparse to take full advantage of the block structure of frame of video block, and wavelet transformation different estate
The hierarchical structure of interband is sparse, realizes low frequency sub-band and the high-frequency sub-band mapping of multi-scale dictionary dictionary pair；On the other hand,
In view of the important function that online dictionary learning plays during training dictionary pair, the method that the present invention passes through online dictionary learning
Training set in each direction of each subband is individually learnt to obtain corresponding base so that obtain it is multiple dimensioned in each
The complete dictionary base of rarefaction representation of yardstick, so enables to the adaptable rarefaction representation of frame block signal, and the sparse table
Show with expression accuracy that is structural, and then improving natural sign in video, moreover it is possible to accelerate the convergence of convex loose restructing algorithm
And stability, contribute to the lifting of the performance and practicality of gradable video encoding of the present invention.
Compared with prior art, the present invention has following beneficial effect：
The present invention substantially increases reconstruction property, with traditional video pressure being reconstructed using fixed base or wavelet basis
Contracting sensor-based system is compared, due to the present invention reconstruct using the global optimum of adaptability gene this on quality reconstruction
It can be strengthened；For other high dimensional signals, the present invention can also be used by appropriate modification, have stronger adaptability；
When rebuilding due to the special tectonic of multi-scale dictionary base so that signal has structural rarefaction representation, therefore the present invention exists
Reconstruction quality can be further improved in the case of identical code rate, while also possesses good scalability.
Brief description of the drawings
The detailed description made by reading with reference to the following drawings to non-limiting example, further feature of the invention,
Objects and advantages will become more apparent upon：
Fig. 1 is one embodiment of the invention based on the sparse multiple dimensioned training set constructing module of stratification and online dictionary
The structured flowchart of study module；
Fig. 2 is the structured flowchart of across the yardstick frame of video reconstructed module of one embodiment of the invention；
Fig. 3 is across the yardstick frame of video reconstructed module of one embodiment of the invention coefficient extension schematic diagram yardstick.
Embodiment
With reference to specific embodiment, the present invention is described in detail.Following examples will be helpful to the technology of this area
Personnel further understand the present invention, but the invention is not limited in any way.It should be pointed out that the ordinary skill to this area
For personnel, without departing from the inventive concept of the premise, various modifications and improvements can be made.These belong to the present invention
Protection domain.
As shown in Figure 1 and Figure 2, the present embodiment provides a kind of gradable video encoding based on multiple dimensioned online dictionary learning
System, including：Based on the sparse multiple dimensioned training set constructing module of stratification, online dictionary learning module and across yardstick video
Frame reconstructed module, wherein：
The multiple dimensioned training set constructing module sparse based on stratification obtains image difference using multi-level wavelet transform
Stratification sparsity structure on yardstick, the elementary area in oriented energy acquisition image is extracted by Difference of Gaussian filter group：To elementary area extraction 7*7 image block, in formulaWithIt is yardstick σ
With the single order on the θ of direction and second order Gauss difference filter, primitive image block generates multiple dimensioned training set；
The online dictionary learning module ensure that the iteration optimization dictionary under low complex degree using stochastic gradient descent method
Atom, dictionary learning, multiple dimensioned sub- word corresponding to generation are carried out to the training set of different scale by online dictionary learning algorithm
Allusion quotation base；
Across the yardstick frame of video reconstructed module arrives different levels to low-frequency video frame by the sub- dictionary base study of construction
Loss high-frequency information, reconstructed by the wavelet inverse transformations of different series, realize the gradable purpose of video quality.
It is described based on the sparse multiple dimensioned training set constructing module of stratification and online dictionary learning in the present embodiment
Module selects a subset as key frame, K is to key frame in decoding end as shown in figure 1, frame of video for one section of sequence
Rank wavelet transformation, the high-frequency sub-band on a low frequency sub-band and (K-1) * 3 directions is obtained, obtained by wavelet inverse transformation different
The wavelet reconstruction signal of exponent number, successively the reconstruction signal with preceding single order subtract each other, obtain reconstructing the high-frequency information of acquisition on each layer
Gain.Edge is carried out by Gaussian filter to the same position for corresponding to low frequency sub-band in the high-frequency information gain on each layer
Primitive blocks are obtained with skin texture detection, obtain multiple dimensioned training setEach sub- training setIt is corresponding
The block structure for the high-frequency information that i-th rank wavelet inverse transformation is obtained, can corresponding to the high-frequency information of low frequency sub-band block structure
Pass through the atom linear expression in corresponding high frequency training set.
In the present embodiment, described online dictionary learning module is in order to adaptive from the training focusing study of large sample to one
The complete dictionary base answered, expression image/video block that can be sparse in acceptable error.Corresponding training set x=[x1, x2...,
xn], it is traditional that experience cost function is reduced by iteration based on batch gradient descent methodOptimization
The dictionary base that dictionary atom, wherein D represent for signal, cost functionTable
Show the degree of error of sparse coding, α is rarefaction representation coefficient, but the method for being all based on whole training set in this iteration every time has
Very high computation complexity and space occupancy rate.Online dictionary by randomly choosing a sample block come excellent in each iteration
Change approximate expectation cost functionDictionary atom is updated based on stochastic gradient descent methodφtFor learning rate,To seek local derviation to D, reduce computation complexity and space accounts for
With rate, while can prove approximate in the case where sample number is sufficiently large it is expected that cost function converges to 0.Its sparse coding algorithm
Realized by LARS algorithms, dictionary atomic update process is by block coordinate gradient descent method (block-coordinate gradient
Descent) realize.
As shown in figure 3, the intersubband of wavelet transformation has the ductility of coefficient, therefore corresponding sit is considered when constructing dictionary
The height of cursor position-low-frequency image block can be approached upper identical rarefaction representation with based on dictionary.Therefore by low
The training set of frequency subband is learnt, and obtains low frequency dictionary base and training set in l optimal thereon1The sparse table of Norm minimum
Show factor alphaL, by convex relaxed algorithm model, α is multiplied by with corresponding training setLObtain excessively complete by dictionary on corresponding yardsticks at different levels
Base.
As shown in Fig. 2 across the yardstick frame of video reconstructed module is by a kind of convex relaxed algorithm model realization, tool
Body is：For the low resolution non-key frame of down-sampling interpolation again, first pass through one-level wavelet decomposition and obtain the small of its low frequency sub-band
Wave system number, pass through the primitive in Difference of Gaussian filter group extraction oriented energy acquisition low resolution key frame as during study
Region, the image block of elementary area is intercepted, the image block of elementary area is obtained in study institute with orthogonal matching pursuit (OMP) algorithm
The l on low frequency dictionary base obtained1The optimal rarefaction representation coefficient of Norm minimumIt is adaptive according to network bandwidth and client
Demand determines the video quality needed for reconstruct, that is, determines the series L needed for wavelet inverse transformation reconstruct, it is desirable to it is non-to reconstruct low resolution
The small echo high-frequency sub-band information of the 1-L layers of key frame, the non-key frame block signal of required reconstruct is obtained by wavelet inverse transformation,
Realize the purpose of quality scalability.
The part being not particularly illustrated in above example of the present invention, it can be realized using prior art.
Implementation result
Key parameter is arranged in the present embodiment：Experiment derives from foreman_cif.yuv with video sequence,
Akiyo.yuv, (the 4 of 352x288:2:The YUV files of 0 form), 48 frames are taken altogether.Every 16 frame is a frame group, is chosen per frame
First three frame of group and first three frame of next frame group are key frame, and the frame of residue 13 of this frame group is non-key frame, the selection of dimension of block
For 7 × 7 pixels.Because the gray-scale map of signal has concentrated most energy, what test was mainly completed on gray-scale map.We
Compare the method H.264/SVC that HHI is proposed.Online dictionary learning method used in the present invention have chosen iterations and be
150 times.
Compared with H.264/SVC, when code check is 240.64kbps, the present embodiment system is averaged foreman.yuv
0.4dB or so reconstruct gain；When code check is 135.34kbps, the present embodiment system obtains average 0.2dB or so reconstruct
Gain.For akiyo.yuv when code check is 170.15kbps, the present embodiment system obtains average 0.3dB or so reconstruct gain；
When code check is 112.04kbps, the present embodiment system obtains average 0.1dB or so reconstruct gain.
Experiment shows that the video sequence that the present embodiment system reconstructing comes out is better than H.264/SVC obtaining on reconstruction quality
Video sequence.
The specific embodiment of the present invention is described above.It is to be appreciated that the invention is not limited in above-mentioned
Particular implementation, those skilled in the art can make various deformations or amendments within the scope of the claims, this not shadow
Ring the substantive content of the present invention.
Claims (6)
1. a kind of gradable video encoding system based on multiple dimensioned online dictionary learning, it is characterised in that the system includes：
Mould is reconstructed based on the sparse multiple dimensioned training set constructing module of stratification, online dictionary learning module and across yardstick frame of video
Block, wherein：
The multiple dimensioned training set constructing module sparse based on stratification obtains image different scale using multi-level wavelet transform
On stratification sparsity structure, pass through Difference of Gaussian filter group extract oriented energy obtain image in elementary area, interception
The image block of elementary area generates multiple dimensioned training set；The multiple dimensioned training set constructing module sparse based on stratification is realized
By K-1 ranks high frequency on the low frequency sub-band and 3 directions of picture is obtained by K rank wavelet transformations on the key frame of reconstruct
Band, the extraction of primitive blocks is carried out by Gaussian filter for each subband and direction is classified, each direction of different scale
Primitive blocks in class correspond to a sub- training set, while have stratification dilute on the dictionary base for training on sub- training set to obtain
Dredge structure；
The online dictionary learning module ensure that the iteration optimization dictionary atom under low complex degree using stochastic gradient descent method,
Dictionary learning, multiple dimensioned sub- dictionary base corresponding to generation are carried out to the training set of different scale by online dictionary learning algorithm；
The online dictionary learning module realizes the minimum that rarefaction representation error is realized by stochastic gradient descent method, its property of can adapt to
The immanent structure for representing high dimensional signal, relative to fixed base can more effectively rarefaction representation vision signal, it is this excessively complete
Rarefaction representation on standby study dictionary basic matrix has structural sparse；
Across the yardstick frame of video reconstructed module learns losing to different levels to low-frequency video frame by the sub- dictionary base of construction
High-frequency information is lost, is reconstructed by the wavelet inverse transformation of different series, realizes the gradable purpose of video quality；
Described online dictionary learning module can be based only upon current training block in each iteration and minimize cost function, single
Solely act on different sub- training set groups and obtain different sub- dictionaries pair；
Described online dictionary learning module is in order to from the training focusing study of large sample to an adaptive complete dictionary
Base, expression image/video block that can be sparse in acceptable error, by randomly choosing a sample in each iteration
Block it is expected cost function to optimize approximationDictionary atom is updated based on stochastic gradient descent methodφtFor learning rate,To seek local derviation to D, reduce computation complexity and space accounts for
With rate；Wherein：Corresponding training set x=[x1, x2..., xn], it is traditional that experience is reduced by iteration based on batch gradient descent method
Cost functionOptimize dictionary atom, the dictionary base that wherein D represents for signal, cost functionThe degree of error of sparse coding is represented, α is rarefaction representation coefficient.
2. a kind of gradable video encoding system based on multiple dimensioned online dictionary learning according to claim 1, it is special
Sign is that the multiple dimensioned training set construction based on wavelet transformation passes through wavelet transformation, primitive blocks extraction and sorting technique
Whole reconstructed key-frame is operable to realize.
3. a kind of gradable video encoding system based on multiple dimensioned online dictionary learning according to claim 2, it is special
Sign is, the multiple dimensioned training set constructing module sparse based on stratification, for the frame of video of one section of sequence, selects one
Individual subset does K rank wavelet transformations to key frame in decoding end, obtains a low frequency sub-band and (K-1) * 3 sides as key frame
Upward high-frequency sub-band, the wavelet reconstruction signal of different rank is obtained by wavelet inverse transformation, the reconstruct with preceding single order successively is believed
Number subtract each other, obtain reconstructing the high-frequency information gain of acquisition on each layer；It is low to corresponding in the high-frequency information gain on each layer
The same position of frequency subband carries out edge by Gaussian filter and skin texture detection obtains primitive blocks, obtains multiple dimensioned training setEach sub- training setThe structure for the high-frequency information that corresponding i-th rank wavelet inverse transformation is obtained
Block, the atom linear expression in corresponding high frequency training set can be passed through corresponding to the high-frequency information of low frequency sub-band block structure.
A kind of 4. gradable video encoding system based on multiple dimensioned online dictionary learning according to claim any one of 1-3
System, it is characterised in that in the online dictionary learning module, sparse coding algorithm is realized by LARS algorithms, dictionary atomic update
Process is realized by block coordinate gradient descent method.
A kind of 5. gradable video encoding system based on multiple dimensioned online dictionary learning according to claim any one of 1-3
System, it is characterised in that across the yardstick frame of video reconstructed module is found most by a kind of convex relaxed algorithm model realization
Excellent rarefaction representation is sparse to be multiplied by corresponding sub- dictionary base, seeks to obtain gradable weight by the wavelet inverse transformation of different rank
Structure signal.
6. a kind of gradable video encoding system based on multiple dimensioned online dictionary learning according to claim 5, it is special
Sign is, across the yardstick frame of video reconstructed module, for the low resolution non-key frame of down-sampling interpolation again, first passes through one
Level wavelet decomposition obtains the wavelet coefficient of its low frequency sub-band, and direction energy is extracted by Difference of Gaussian filter group as during study
Amount obtains the elementary area in low resolution key frame, intercepts the image block of elementary area, is obtained with orthogonal matching pursuit algorithm
The image block of elementary area is on the low frequency dictionary base obtained by studyThe optimal rarefaction representation coefficient of Norm minimumIt is adaptive
The video quality according to needed for network bandwidth and client demand determine reconstruct answered, that is, determine needed for wavelet inverse transformation reconstruct
Series L, it is desirable to reconstruct the small echo high-frequency sub-band information of the 1-L layers of low resolution non-key frame, obtained by wavelet inverse transformation
The non-key frame block signal of required reconstruct, realize the purpose of quality scalability.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410331199.6A CN104199627B (en) | 2014-07-11 | 2014-07-11 | Gradable video encoding system based on multiple dimensioned online dictionary learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410331199.6A CN104199627B (en) | 2014-07-11 | 2014-07-11 | Gradable video encoding system based on multiple dimensioned online dictionary learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104199627A CN104199627A (en) | 2014-12-10 |
| CN104199627B true CN104199627B (en) | 2018-01-19 |
Family
ID=52084925
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410331199.6A Active CN104199627B (en) | 2014-07-11 | 2014-07-11 | Gradable video encoding system based on multiple dimensioned online dictionary learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104199627B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104539961B (en) * | 2014-12-12 | 2017-12-15 | 上海交通大学 | Gradable video encoding system based on the gradual dictionary learning of hierarchy |
| CN104820842A (en) * | 2015-05-19 | 2015-08-05 | 重庆大学 | Mining and visualization method for supply and distribution energy consumption data in big-data environment |
| CN105163121B (en) * | 2015-08-24 | 2018-04-17 | 西安电子科技大学 | Big compression ratio satellite remote sensing images compression method based on depth autoencoder network |
| CN105721869B (en) * | 2016-01-26 | 2018-04-06 | 上海交通大学 | The collection of compression tensor and reconfiguration system based on structural sparse |
| CN108319888B (en) * | 2017-01-17 | 2023-04-07 | 阿里巴巴集团控股有限公司 | Video type identification method and device and computer terminal |
| CN111161128A (en) * | 2019-11-18 | 2020-05-15 | 田树耀 | Image transformation based on frequency domain direction filtering and application thereof in sparse decomposition |
| CN111369443B (en) * | 2020-03-19 | 2023-04-28 | 浙江昕微电子科技有限公司 | Zero-order learning super-resolution method of light field cross-scale |
| CN111582048B (en) * | 2020-04-16 | 2022-09-23 | 昆明理工大学 | Undersampled signal high-resolution reconstruction method based on dictionary learning and sparse representation |
| CN112184840B (en) | 2020-09-22 | 2022-09-02 | 上海交通大学 | 3D point cloud compression system based on multi-scale structured dictionary learning |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102427527A (en) * | 2011-09-27 | 2012-04-25 | 西安电子科技大学 | Method for reconstructing non key frame on basis of distributed video compression sensing system |
| CN103618907A (en) * | 2013-11-08 | 2014-03-05 | 天津大学 | Multi-viewpoint distributed type video encoding and frame arranging device and method based on compressed sensing |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8179976B2 (en) * | 2008-01-11 | 2012-05-15 | Apple Inc. | Control of video decoder for reverse playback operation |
-
2014
- 2014-07-11 CN CN201410331199.6A patent/CN104199627B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102427527A (en) * | 2011-09-27 | 2012-04-25 | 西安电子科技大学 | Method for reconstructing non key frame on basis of distributed video compression sensing system |
| CN103618907A (en) * | 2013-11-08 | 2014-03-05 | 天津大学 | Multi-viewpoint distributed type video encoding and frame arranging device and method based on compressed sensing |
Non-Patent Citations (2)
| Title |
|---|
| Multiscale Online Dictionary Learning for Quality Scalable Video Coding;Xin Tang etc.;《Data Compression Conference》;20140605;第428页第1-29行 * |
| STOL:Spatio-Temporal Online Dictionary Learning for low bit-rate wideo coding;Xin Tang etc.;《Data Compression Conference》;20130620;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104199627A (en) | 2014-12-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104199627B (en) | Gradable video encoding system based on multiple dimensioned online dictionary learning | |
| Shi et al. | Image compressed sensing using convolutional neural network | |
| CN110119780B (en) | Hyper-spectral image super-resolution reconstruction method based on generation countermeasure network | |
| Zhang et al. | Image compressive sensing recovery via collaborative sparsity | |
| Li et al. | An efficient deep convolutional neural networks model for compressed image deblocking | |
| Zhang et al. | CCR: Clustering and collaborative representation for fast single image super-resolution | |
| CN102123278B (en) | Signal source encoding and decoding method based on distributed compressive sensing technology | |
| Liu et al. | Data-driven soft decoding of compressed images in dual transform-pixel domain | |
| CN104574336B (en) | Super-resolution image reconstruction system based on adaptive sub- mould dictionary selection | |
| CN102708576A (en) | Method for reconstructing partitioned images by compressive sensing on the basis of structural dictionaries | |
| Zhao et al. | Image super-resolution via adaptive sparse representation | |
| Zhao et al. | CREAM: CNN-REgularized ADMM framework for compressive-sensed image reconstruction | |
| CN113516601A (en) | Image restoration technology based on deep convolutional neural network and compressed sensing | |
| CN109949217A (en) | Video super-resolution method for reconstructing based on residual error study and implicit motion compensation | |
| Zebang et al. | Densely connected AutoEncoders for image compression | |
| CN110288529B (en) | Single image super-resolution reconstruction method based on recursive local synthesis network | |
| CN114449276B (en) | Super prior side information compensation image compression method based on learning | |
| Pham et al. | CSIE-M: compressive sensing image enhancement using multiple reconstructed signals for internet of things surveillance systems | |
| Feng et al. | Sub clustering K-SVD: size variable dictionary learning for sparse representations | |
| Lv et al. | Hoca: Higher-order channel attention for single image super-resolution | |
| Wang et al. | Image reconstruction from patch compressive sensing measurements | |
| Wang et al. | A customized deep network based encryption-then-lossy-compression scheme of color images achieving arbitrary compression ratios | |
| CN109819258A (en) | A kind of splits' positions perceived direction predictive coding method based on spiral scan | |
| Emara et al. | Image compression using advanced optimization algorithms | |
| CN102833537A (en) | Compressed sensing image sparse method based on Contourlet transformation |
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 |