CN104094607B - Modeling method and system based on context in transform domain of image/video - Google Patents
Modeling method and system based on context in transform domain of image/video Download PDFInfo
- Publication number
- CN104094607B CN104094607B CN201280027747.5A CN201280027747A CN104094607B CN 104094607 B CN104094607 B CN 104094607B CN 201280027747 A CN201280027747 A CN 201280027747A CN 104094607 B CN104094607 B CN 104094607B
- Authority
- CN
- China
- Prior art keywords
- coefficient
- state
- image
- block
- context
- 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.)
- Expired - Fee Related
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
- H04N19/10—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
- H04N19/12—Selection from among a plurality of transforms or standards, e.g. selection between discrete cosine transform [DCT] and sub-band transform or selection between H.263 and H.264
- H04N19/122—Selection of transform size, e.g. 8x8 or 2x4x8 DCT; Selection of sub-band transforms of varying structure or type
Landscapes
- Physics & Mathematics (AREA)
- Discrete Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
A modeling method and system based on the context in the transform domain of an image/video in the image/video coding/decoding and other processing technical fields. Compression or other processing can be performed on dynamic or static images by approximating a natural image to a two-dimensional Markov random field which has radial and reverse radial relevance simultaneously in the DCT transform domain or another transform domain and depicting the direction relevance of an image signal in a two-dimensional change domain using the model.
Description
Technical field
The present invention relates to the method and system of a kind of image/video coding decoding and other processing technology fields, tool
Body is the two dimension that one kind portrays natural image anisotropy dependency in a certain transform domain (such as discrete cosine transform (DCT) domain)
The modeling method and its system of Markov random field (Markov random field).
Background technology
Current image/video treatment technology, such as compression, reconstruct, enhancing, analysis etc., are carried out in transform domain mostly.
Conventional transform domain has discrete cosine transform (DCT), discrete Fourier transform (DFT) (DFT), Hadamard transform etc..The effect of conversion is
Among energy in image/video signals is concentrated on minority conversion coefficient, so as to significantly reduce the statistical redundancy in signal.
Academia has various descriptions to the process：Such as energy packing (energy packing), decorrelation or image/video signals is dilute
Relieving the exterior syndrome shows.Even if in the transform domain as illustrated, natural image/video signal is also far from independent same distribution (i.i.d), and should be returned
Class is in Markov random process (Markov random processes).Therefore, for image/video signals are in transform domain
In context statistical modeling become the key components in the image/video processing system being widely used.Here it is so-called
Context statistical modeling refer to markov or approximate markov signal conditional probability method of estimation or process.
Based on conversion image/video compression system, such as MPEG, JPEG, JPEG2000, H.264 in, image/video
The context statistical modeling of signal is undoubtedly most important to rate distortion (rate-distortion) performance of system, and its effect is
Predict for driving the conditional probability of the conversion coefficient of entropy coder (such as context-based arithmetic coding device etc.).In entropy code
During, any prediction deviation for conditional probability can all directly result in the decline of coding efficiency.Precisely, with theory most
The long ratio of short code, the redundancy code length caused by prediction probability deviation are equal to which with the Mutual information entropy between true probability distribution
(relative entropy), or KL distances (Kullback-Leibler distance).Therefore context statistical modeling
Precision finally determines the compression performance of system.
Existing context modeling method often has the characteristic of the power spectrum of rapid decrease (in document using natural image
It is assumed to be index decreased).But the power spectrum with rapid decrease itself cannot characterize image/video signals system in the transform domain as illustrated
Meter characteristic, this is because the signal energy of image and video is only one-dimensional distribution on frequency domain, and is then two dimension in the transform domain as illustrated
Or it is three-dimensional.
Especially such as border, texture etc. is containing directive image with passing through its conversion coefficient body in the transform domain of two dimension
The characteristics of revealing two-dimensional directional dependency, the directional dependency of above-mentioned conversion coefficient are being carried out to the block of image pixels containing border
Two-dimensional discrete is especially pronounced when converting, and for the image block containing border or regular veins, signal energy is focused into direction
In subband, shown in such as Fig. 1 (a) and 1 (c), as discrete cosine transform is carried out comprising the different block of image pixels towards border
(DCT) result.
For the block of pixels comprising smooth shade, its signal energy then concentrates on low frequency region, and dct transform coefficient is with spoke
Shape and the decay of intimate same speed are penetrated, shown in such as Fig. 1 (b).It can be seen that natural image can in discrete cosine transform (DCT) domain
To be approximately while there is radially (radial direction) and against radially (anti-radial direction) dependency
Two-dimensional Markov random field models.
H.264 various existing International image video compression standards are disclosed by above-mentioned analysis, such as JPEG, MPEG and the institute such as
Using common DCT coefficient (Zigzag) scan method inherently has defect in a zigzag.Zigzag scannings are inverse radial direction
(anti-radial direction) reciprocating scanning, thoroughly ignores the statistic correlation that natural image diametrically has.
In fact, MPEG and H.264 standard are in order to make up the defect, scan pattern conduct both horizontally and vertically is proposed respectively
The replacement of Zigzag scan modes.But the switching of many scan patterns be only local and inflexible tentative plan, it is impossible to image/
Video signal any direction dependency in the transform domain as illustrated is modeled, and the coding of scan pattern will also result in extra
Code check expense.
Find through the retrieval to prior art, Chinese patent literature CN1741616, publication date 2006-03-01, note
A kind of adaptive entropy coding method based on context is carried, the technology is comprised the following steps：During coding：Scanning Current Transform block
In the DCT coefficient that has been quantized, be consequently formed (level, run) several to sequence;Then by the reverse order logarithm of scanning to sequence
Each several to carrying out entropy code in row, in coding, the value dynamic using be encoded in block having completed coding several pairs is adaptive
Context statistical model should be constructed, while, it was also proposed that a context model Weighted Fusion technology is further improving model
Compression performance;The context statistical model obtained with previous step is driving entropy code.Adaptive entropy solution based on context
Code method is the inverse of coded method.But the technology has following defect and deficiency：Although the method adopts the distance of swimming (run-
Length mode) has carried out assembly coding, but the method not to others to continuous 0 coefficient between two non-zero coefficients
Typical coefficient combination is merged, and still fails to jump out the simple scanning mode of Zigzag.
Chinese patent literature CN1431828, publication date 2003-07-23 are described a kind of " for encoding/decoding image
The optimum scanning method of signal ", the technology in a kind of method by discrete cosine transform coding picture signal, in multiple ginsengs
Examine in block at least one to be chosen.A scanning sequency is produced, wherein the block to be encoded of reference block is scanned, and with institute
The scanning sequency of generation scans block to be encoded.At least one selected reference block is with block to be encoded in time or sky
Between it is upper neighbouring.When block to be encoded is scanned, generation nonzero coefficient is obtained from described at least one selected reference block
Probability, and start to determine scanning sequency with descending from highest probability.Here, if probability were identical, scanning sequency quilt
It is produced as a flexuose scanning sequency.But the method still fails the drawbacks of fundamentally solving Zigzag scan patterns,
And its adaptive scanning to conversion coefficient is still for single coefficient to carry out, to possessing approximate property it is not
Several piece is merged, therefore, the code efficiency of the method is still not enough to satisfactory.
The content of the invention
The present invention is directed to deficiencies of the prior art, there is provided based on upper and lower in a kind of transform domain of image/video
The modeling method and its system of text, by adaptive block Evolve-ment law (ABE：Adaptive Block Evolution), it is and existing
Serration type (Zigzag), scan mode horizontally or vertically are different, and the method does not adopt fixed one-dimensional scanning order, but
Using self-adaption two-dimensional scan mode, statistical modeling is carried out to dependency of the coefficient in transform domain simultaneously from radial direction and inverse radial direction.
The present invention is in addition to can be as the effective tool in image/video compression, additionally it is possible to for performing denoising, inserting
Other image/videos such as value, classification, visual information retrieval and extraction, digital watermarking, Information hiding, image retrieval, steganalysis
Process in application.
The present invention is achieved by the following technical solutions：
The present invention relates to a kind of context statistical modeling method in transform domain, by adaptively building a two dimension Ma Er
Can husband's model to react directional dependency of the image/video signals in two-dimentional domain of variation, the method is specifically included：
Multiple or single conversion coefficient is constituted into the state in two-dimensional Markov process；
The transfer probability between two adjacent Markovian states is calculated online or offline；
From original state, the self adaptation of single or multiple traversal directions is realized respectively by transfer probability.
Described conversion coefficient can be through quantization or without quantization；
State in described two-dimensional Markov process refer to or a group in transform domain it is adjacent and related be
Number, the state are defined by the coefficient block of typical module；
Described original state is referred to：With low-limit frequency coefficients and/or highest frequency coefficient as reference point, flock together
With same or like numerical value and close frequencies the coefficient block that constituted of coefficient group；The intercontinental of the coefficient group is broad sense half
Footpath.
Described transfer probability P (si|si-1) transfer probability that refers exclusively in two-dimensional Markov process, can by offline or
The online mode of person is calculated, wherein：siFor NextState, si-1For current state, i represents the conversion coefficient of current traversal
Coefficient block sequence number, its value is the state sum in 1 to two-dimensional Markov process.
The self adaptation of described traversal direction is referred to：According to the biography between current state and its all next possible state
Pass state corresponding to the peak in probability to select next state；For plural original state is then carried out respectively
Above-mentioned adaptive polo placement.
The present invention relates to a kind of image/video compression method based on transform domain context statistical modeling, including following step
Suddenly：
The first step, transformation of coefficient is carried out for input picture；
Second step, the modeling based on context is carried out to conversion coefficient and original state s is determined0；
It is 3rd step, online or offline with two-dimentional markoff process to state si-1Under be possible to transfer probability P
(si|si-1) calculated and compared, and based on optimal value therein as next state si；
4th step, next state s obtained by the 3rd stepi, with its corresponding transfer probability P (si|si-1) drive entropy to compile
Code device, output state siThe output code of corresponding variation coefficient, is then back to the 3rd step and re-starts meter according to new state
Calculate and compare, until having traveled through all conversion coefficients and having obtained the complete code stream of all output code compositions.
Described original state adopt the mode of broad sense radius to encode at least one value in minimum frequency domain for 1 coefficient or be
At least one value in several piece and/or highest frequency domain for 0 coefficient or coefficient block and/or frequency domain arbitrary portion in have it is substantially special
The coefficient block levied.
Described obvious characteristic is referred to：Two or more adjoin and with identical value or the coefficient in exemplary distribution to or coefficient
Block.
The present invention relates to a kind of image/video compression system based on transform domain context statistical modeling, including：Become mold changing
Block, adaptive block evolution (ABE) module and entropy code module, wherein：
Conversion module, carries out transformation of coefficient to image；
ABE modules, carry out the modeling based on context to the conversion coefficient that conversion module is exported, and according to adaptive mode
Progressively the transfer probability that each step in ergodic process is obtained is sequentially output to and entropy code module is driven.
Entropy code module, carries out entropy code and exports according to the transfer probability of ABE modules output to conversion coefficient.
Described is referred to based on the modeling of context：Using in transform domain or one group of adjacent and related coefficient as
The state of two-dimentional markoff process, with low-limit frequency coefficients and/or highest frequency coefficient as reference point, the tool for flocking together
Original state of the coefficient block constituted by the coefficient group for having same or like numerical value and close frequencies as model.
Described adaptive mode is referred to：From the original state based on the model of context, its all next one is calculated
The state corresponding to the peak in transfer probability between possible state is used as next state.
It is of the invention to improve code efficiency compared with existing H.264 compress mode, especially when code check be increased to 0.45 or with
When upper, the code stream length of ABE coding systems can be reduced to being close to 90%.
Description of the drawings
Fig. 1 is the 16x16 sizes block of pixels correspondence dct transform coefficient schematic diagram (coefficient amplitude of different images in embodiment 1
Illustrated by bright-dark degree).
Fig. 2 is the contrast of the ABE coding systems of traditional DCT coding systems and the present invention.
Fig. 3 is the multiple dct transform coefficient block schematic diagrams of 8x8 sizes under Markov state, concrete such as grey area in figure
Domain.
Fig. 4 is the schematic diagram for carrying out state transformation with high pass probability as direction, and the i-th step is walked to the i-th v；
In figure：A () is radial direction state change；B () changes for horizontality.
One-dimensional or two-dimensional directional transverse scan schematic diagrams of the Fig. 5 for transform domain；
In figure：A () is to be traveled through from low transform domain as original state；B () is as original state from high transform domain
Traveled through；C () is for respectively from low transform domain and high transform domain while traveled through.
Fig. 6 is image set used by the test experiments for verifying coding efficiency of the present invention.
Fig. 7 is coding efficiency of the present invention comparing result with existing best coding system H.264.
Specific embodiment
Below embodiments of the invention are elaborated, the present embodiment is carried out under premised on technical solution of the present invention
Implement, give detailed embodiment and specific operating process, but protection scope of the present invention is not limited to following enforcements
Example.
1 two-dimentional Markov modeling of embodiment
Further to embody the novelty of ABE modes of the present invention, below with numerous image/video compression standards and system
In core the most processing procedure, that is, as a example by the entropy code of the two dimensional DCT coefficients after quantifying.Traditional DCT coding systems and sheet
Invention ABE coding system block diagrams are as shown in Figure 2.Similar with traditional DCT coding systems, also input picture is carried out ABE systems first
Dct transform and quantization, but it is different from the zigzag scanning (Zigzag) adopted by legacy system, and ABE systems are using more flexible
Adaptive block evolvement method, so as to possess higher code efficiency.
After quantified, most of non-significant DCT coefficient is quantified as 0, and remaining non-zero, i.e., notable coefficient, often edge
Gather in certain direction.Contrary substantial amounts of non-significant coefficient (0), then can be along the fan-shaped diffusion in direction from low to high.To improve
Compression efficiency, ABE methods adopt orderly planar Marcov model and adjacent pixel blocks sequence number to decompose 0 and non-zero position (ratio
Such as：Generate weight map).Will be to current state s in above-mentioned two-dimensional Markov modeling processi-1To next state siTransmission
Probability P (si|si-1) be predicted.Described state refers to one group of adjacent and related coefficient block in transform domain.Such as one group
Full 0 coefficient block positioned at highest frequency domain constitutes a state (such as Fig. 4 (a))；And another group of complete non-zero coefficient positioned at minimum frequency domain
Block then constitutes another state (such as Fig. 4 (b)).More generally useful, state is then by 0 coefficient of part and non-zero coefficient optionally structure
Into (such as Fig. 3 (c)).
Used as entropy code is carried out to the transform domain DCT coefficient after quantization, ABE methods pass through si-1→siThe conversion of mode with
Improve transfer probability P (si|si-1), realization is traversed the mode of transform domain and is scanned.Original state s in conversion process0May
It is one and is located at high-frequency domain and comprising 0 big coefficient block, or one is located at lower frequency region and comprising non-zero big coefficient block, point
Not as shown in Fig. 4 (b) and Fig. 4 (a).These full 0s or complete 1 original state can be conveniently by the broad sense of an integer numerical value half
Coding is realized in footpath, and the method for the broad sense radius more effectively and is widely used in existing compression standard compared with block tail (EOB) method.
The intermediateness traversed in formula scanning process in two-dimensional transform domain is by continuous 0 positioned at Current Scan region header
Coefficient or non-zero coefficient composition.Context statistical modeling is carried out by the conversion to adjacent states, ABE methods are realized to transform domain
Particular orientation (most of for radially) on dependency utilization.
As the directivity of conversion coefficient is closely related with the direction character in image block, when two adjacent states siWith
si-1When being same type and identical direction, its transfer probability P (si|si-1) get Geng Gao will be showed.By using the rule,
ABE methods can realize adaptive direction state transformation in formula scanning process is traversed or during compressed transform coefficient, such as Fig. 4 institutes
Show.
The direction of propulsion traveled through to conversion coefficient can be unidirectional：From high frequency to low frequency or from low frequency to height
Frequency finally converges to intermediate frequency propulsion from high and low frequency respectively but it is also possible to be two-way.Fig. 5 is these three traversal modes
Signal.
By reasonably select Markovian state and implement high probability conversion, ABE methods can with based on context
Matching entropy coder combines, and conversion coefficient is encoded with shorter code length.
2 compression of images application of embodiment
The present embodiment is comprised the following steps：
The first step, transformation of coefficient is carried out for input picture；
Described transformation of coefficient can adopt dct transform or KLT (Karhunen-Loeve) to convert, or using existing known
Other alternative approachs are combined with following traversal mode, Jing experiment examine can reach with shown in the present embodiment and Fig. 7
Akin effect.
Described transformation of coefficient can be processed with additional quantization；
Second step, the modeling based on context is carried out to conversion coefficient and original state s is determined0；
Original state in the present embodiment is encoded by the way of broad sense radius, and the original state can be following any one
Kind：
I) some numerical value for being gathered in minimum frequency domain are 1 coefficient block, shown in such as Fig. 5 (a)；
Ii some numerical value for) being gathered in highest frequency domain are 0 coefficient block, shown in such as Fig. 5 (b)；
Iii) it is above-mentioned i) and ii) combination, shown in such as Fig. 5 (c).
It is 3rd step, online or offline with two-dimentional markoff process to current state si-1Be possible to transfer probability P
(si|si-1) calculated and compared, and based on maximum therein as next state si。
4th step, next state s obtained by the 3rd stepi, with transfer probability P (si|si-1) go to drive entropy coder, it is defeated
Do well siThe output code of corresponding variation coefficient, is then back to the 3rd step and re-starts calculating according to new state and compare
Compared with until having traveled through all conversion coefficients and having obtained the complete code stream of all output code compositions.
In above-mentioned steps, i is the state sum in 1 to two-dimensional Markov process；As original state number more than one,
Can then adopt as traveled through to improve coding rate while the mode of Fig. 5 (c) is realized on two or more direction.
In this application, be related to said method realizes system, including：Conversion module, adaptive block evolution (ABE) module
And entropy code module, wherein：
Conversion module, carries out the transformation of coefficient such as DCT or KLT to image；
ABE modules, carry out the modeling based on context to the conversion coefficient that conversion module is exported, and according to adaptive mode
Highest transition probability P (s are selected progressivelyi|si-1) corresponding statess carry out redirecting for markoff process, finally travel through all coefficients,
Simultaneously by P (s in ergodic processi|si-1) be sequentially output to entropy code module to drive entropy coder.
Entropy code module, carries out entropy code according to the transfer probability that ABE modules are progressively exported and exports to conversion coefficient.
Described conversion module can be accompanied with quantification treatment function；
Described is referred to based on the modeling of context：Using in transform domain or one group of adjacent and related coefficient as
The state of two-dimentional markoff process, with low-limit frequency coefficients and/or highest frequency coefficient as reference point, the tool for flocking together
Original state of the coefficient block constituted by the coefficient group for having same or like numerical value and close frequencies as model.
Described adaptive mode is referred to：From the original state based on the model of context, its all next one is calculated
Transfer probability between possible state, and next state is selected based on peak therein.
The present embodiment verifies the volume of ABE systems described in embodiment 1 using the conventional test image of 38 width as shown in Figure 6
Code performance.Each image first passes around the dct transform of 8x8, then using different quantization steps respectively to the notable of DCT coefficient
Figure (significant map) is encoded.As a comparison, we are from the best H.264 encoder of current effect, to phase
With notable figure being encoded, compare two methods in order to objective, entropy code module using H.264 encoder acquiescence from
Adapt to binary arithmetic encoder CABAC.Also, the probability of all contexts is both configured to 0.5.It should be noted that
In this example, the context that ABE coding systems are adopted（context）Number is three times of (378vs126 H.264）, so ABE
The context dilution of system（context dilution）Punishment is even more serious.Therefore, the Initialize installation mode more has in fact
Beneficial to H.264 coding system.
We compare the relative ratio of final the generated code stream length of two kinds of coded systems：γ=(L264-LABE)/L264, its
In, LABERepresent the code stream length that ABE coding systems are generated, L264Represent the code stream length that H.264 system is generated.Different images
γ under different code checks as shown in fig. 7, it can be seen that compared with H.264, ABE coding systems are clearly more efficient,
Especially when code check is raised, the advantage of ABE coding systems becomes apparent from.Additionally, test indicate that, ABE systems are to other coefficients
Conversion, such as KLT conversion etc., and other existing entropy coders are adopted, such as the MQ encoders adopted by JPEG2000, and Hough
Graceful encoder etc., is all up the performance boost similar to Fig. 7.
Claims (7)
1. in a kind of transform domain context statistical modeling image/video compression method, it is characterised in that comprise the following steps：
The first step, transformation of coefficient is carried out for input picture；
Second step, the modeling based on context is carried out to conversion coefficient and original state s is determined0；
3rd step, in current state si-1Under, with the transfer probability P (s of the online or offline two-dimentional markoff process estimatedi|
si-1) in maximum selection rule NextState si；
4th step, the NextState s to above-mentioned selectioni, with transfer probability P (si|si-1) go to drive entropy coder, output state si
The code stream of corresponding variation coefficient, be then back to the 3rd step recalculate update after state corresponding transfer probability maximum
Value, until having traveled through all conversion coefficients and having obtained complete code stream；
Described is referred to based on the modeling of context：By adaptively building a planar Marcov model to reflect figure
Directional dependency of the picture/video signal in two-dimensional transform domain, specially：Multiple or single conversion coefficient is constituted into two dimension Ma Er
State that can be during husband, i.e. in transform domain or one group of adjacent and related coefficient, the state is by typical module
Coefficient block is defined；
The transfer probability between two adjacent Markovian states, transfer probability P (s are calculated online or offlinei|si-1) refer exclusively to
Transfer probability in two-dimensional Markov process, is calculated by offline or online mode, wherein：siFor next shape
State, si-1For current state, i represents the sequence number of the coefficient block of the conversion coefficient of current traversal, and its value is 1 to two-dimentional Ma Erke
State sum during husband；
From original state, by the comparison to being estimated transfer probability, the adaptive of single or multiple traversal directions is realized respectively
Should, the original state is referred to：With low-limit frequency coefficients and/or highest frequency coefficient as reference point, flock together with phase
With or close numerical value and close frequencies the coefficient block that constituted of coefficient group；The self adaptation of traversal direction is referred to：According to current shape
The state corresponding to the peak in transfer probability between state and its all next possible state is used as next state；It is right
Above-mentioned adaptive polo placement is carried out respectively then in plural original state.
2. image/video compression method according to claim 1, is characterized in that, described original state is using broad sense half
The mode in footpath encodes coefficient or coefficient block and/or at least one value in highest frequency domain that at least one value in minimum frequency domain is 1
Coefficient block with obvious characteristic in coefficient or coefficient block and/or frequency domain arbitrary portion for 0.
3. image/video compression method according to claim 2, is characterized in that, described obvious characteristic is referred to：Two with
On adjoin and with identical value or the coefficient in exemplary distribution to or coefficient block.
4. image/video compression method according to claim 1 and 2, is characterized in that, described original state is following
Meaning is a kind of：
If the dry values for i) being gathered in minimum frequency domain are 1 coefficient block；
Ii some numerical value for) being gathered in highest frequency domain are 0 coefficient block；
Iii) it is above-mentioned i) and ii) combination.
5. a kind of image/video compression system for realizing any of the above-described claim methods described, it is characterised in that include：Become
Mold changing block, adaptive block evolution Markov MBM and entropy code module, wherein：
Conversion module, carries out transformation of coefficient to image；
Adaptive block evolution Markov MBM, carries out building based on context to the conversion coefficient of conversion module output
Mould, and the state of markoff process is progressively built and is traveled through according to adaptive mode；
Entropy code module, carries out entropy to conversion coefficient according to the transfer probability of adaptive block evolution Markov MBM output
Encode and export.
6. image/video compression system according to claim 5, is characterized in that, described based on the modeling of context is
Refer to：Using in transform domain or one group of adjacent and related coefficient as two-dimentional markoff process state, with lowest frequency
Rate coefficient and/or highest frequency coefficient be reference point, flock together with same or like numerical value and close frequencies are
Original state of the several groups of coefficient blocks for being constituted as model.
7. image/video compression system according to claim 6, is characterized in that, described adaptive mode is referred to：From base
In the original state of the model of context, the transfer probability between its all next possible state is calculated, and based on wherein
Peak corresponding to condition selecting the next one state.
Applications Claiming Priority (4)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US201161518246P | 2011-05-04 | 2011-05-04 | |
| US61/518246 | 2011-05-04 | ||
| USUS61/518246 | 2011-05-04 | ||
| PCT/CN2012/075052 WO2012149904A1 (en) | 2011-05-04 | 2012-05-03 | Modeling method and system based on context in transform domain of image/video |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104094607A CN104094607A (en) | 2014-10-08 |
| CN104094607B true CN104094607B (en) | 2017-04-26 |
Family
ID=47107767
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201280027747.5A Expired - Fee Related CN104094607B (en) | 2011-05-04 | 2012-05-03 | Modeling method and system based on context in transform domain of image/video |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN104094607B (en) |
| WO (1) | WO2012149904A1 (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113362252B (en) * | 2021-06-30 | 2024-02-02 | 深圳万兴软件有限公司 | Intelligent picture reconstruction method, device, computer equipment and storage medium |
| CN114822025B (en) * | 2022-04-20 | 2023-04-18 | 合肥工业大学 | Traffic flow combined prediction method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1741616A (en) * | 2005-09-23 | 2006-03-01 | 联合信源数字音视频技术（北京）有限公司 | Adaptive entropy coding/decoding method based on context |
| US7123656B1 (en) * | 2001-10-01 | 2006-10-17 | Realnetworks, Inc. | Systems and methods for video compression |
| CN1874509A (en) * | 2001-09-14 | 2006-12-06 | 诺基亚有限公司 | Method and system for context-based adaptive binary arithmetic coding |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE10218541A1 (en) * | 2001-09-14 | 2003-04-24 | Siemens Ag | Context-adaptive binary arithmetic video coding, e.g. for prediction error matrix spectral coefficients, uses specifically matched context sets based on previously encoded level values |
-
2012
- 2012-05-03 WO PCT/CN2012/075052 patent/WO2012149904A1/en active Application Filing
- 2012-05-03 CN CN201280027747.5A patent/CN104094607B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1874509A (en) * | 2001-09-14 | 2006-12-06 | 诺基亚有限公司 | Method and system for context-based adaptive binary arithmetic coding |
| US7123656B1 (en) * | 2001-10-01 | 2006-10-17 | Realnetworks, Inc. | Systems and methods for video compression |
| CN1741616A (en) * | 2005-09-23 | 2006-03-01 | 联合信源数字音视频技术（北京）有限公司 | Adaptive entropy coding/decoding method based on context |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2012149904A1 (en) | 2012-11-08 |
| CN104094607A (en) | 2014-10-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN1303820C (en) | Quality based image compression | |
| CN105791845B (en) | The method that entropy decoding is carried out to transformation coefficient | |
| CN1189038C (en) | An enbedded DCT-based still image coding algorithm | |
| CN1215720C (en) | Preprocessing method applied to textures of arbitraily shaped objects | |
| CN1200568C (en) | Optimum scanning method for change coefficient in coding/decoding image and video | |
| CN1181690C (en) | Encoding method for compression of video sequence | |
| RU2504103C1 (en) | Method and apparatus for encoding and decoding image using rotational transform | |
| CN103004195B (en) | A variety of variable length decoding methods are used for the video compress of polytype transformation coefficient block | |
| CN1640142B (en) | Method and apparatus for coding of wavelet transformed coefficients | |
| US20130016789A1 (en) | Context modeling techniques for transform coefficient level coding | |
| CN105721878A (en) | Image Processing Device And Method For Intra-Frame Predication In Hevc Video Coding | |
| CN108881922A (en) | Decode the device of Saliency maps | |
| CN1689026A (en) | Context-based adaptive variable length coding for adaptive block transforms | |
| CN107911699A (en) | The coding method of video and equipment and non-transitory computer-readable medium | |
| CN108141593A (en) | For be directed to the efficient intraframe coding of deep video based on the discontinuous method of depth | |
| CN103222263A (en) | Multi-input adaptive filter based on combination of sum-odified laplacian filter indexing and quadtree partitioning | |
| CN101383972A (en) | Remote sensed image compression method based on space prediction and transformation | |
| CN104094607B (en) | Modeling method and system based on context in transform domain of image/video | |
| Xiao et al. | An optimized quantization technique for image compression using discrete Tchebichef transform | |
| KR100359813B1 (en) | Alternative Double Scan For Video Coding And Decoding | |
| CN106911935A (en) | Method of designing integrated circuit based on HEVC entropy coders | |
| CN100426866C (en) | Full-phase Walsh biorthogonal conversion and its improved process for JPEG | |
| CN103561275A (en) | Image compression method applying improved bandelet transformation | |
| CN1720746A (en) | Method of measuring blocking artefacts | |
| Senapati et al. | Low bit rate image compression using hierarchical listless block-tree DTT algorithm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20170327 Address after: Room 6-76, No. 66 academician Road, high tech Zone, Ningbo, Zhejiang, China Applicant after: Ningbo view of the original network technology Co., Ltd. Address before: Shanghai, Minhang District, China No. 800 Dongchuan Road Applicant before: Wu Xiaolin Applicant before: Niu Yi |
|
| TA01 | Transfer of patent application right | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170426 Termination date: 20210503 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |