CN103024399A - Wavelet transform based extreme-low bit-rate video compressing and coding method - Google Patents
Wavelet transform based extreme-low bit-rate video compressing and coding method Download PDFInfo
- Publication number
- CN103024399A CN103024399A CN2013100196008A CN201310019600A CN103024399A CN 103024399 A CN103024399 A CN 103024399A CN 2013100196008 A CN2013100196008 A CN 2013100196008A CN 201310019600 A CN201310019600 A CN 201310019600A CN 103024399 A CN103024399 A CN 103024399A
- Authority
- CN
- China
- Prior art keywords
- coefficient
- wavelet
- decomposition
- subband
- coefficients
- 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.)
- Pending
Links
Images
Abstract
The invention discloses a wavelet transform based extreme-low bit-rate video compressing and coding method. The wavelet transform based extreme-low bit-rate video compressing and coding method includes three steps: (1) subjecting a GOF (group of frame) image group to three-dimensional wavelet decomposition, selecting appropriate decomposition grades for time decomposition and space decomposition according to the size of the image group and formats including CIF (common intermediate format), QCIF (quarter common intermediate format), SIF (standard intermediate format) of videos, decomposing to obtain a coefficient matrix of wavelet decomposition; (2) subjecting the coefficient matrix to branching and blocking operation: firstly subjecting the coefficient matrix to first-grade virtual decomposition or second-grade virtual decomposition, and then blocking the coefficients after virtual decomposition; (3) realizing SPIHT (set partitioning in hierarchical tree) coding process. By the aid of the wavelet transform based extreme-low bit-rate video compressing and coding method, a conventional video wavelet 3D-SPIHT (three-dimensional set partitioning in hierarchical tree) algorithm is improved, a wavelet transformation coefficient is organized in a form of combining a virtual tree with a coefficient branching organization, a multi-threshold sub-band coding optimization method is utilized, important coefficient is coded preferentially, the purpose for displaying more effect information with few bit numbers is achieved, compression ratio can be improved, and compression efficiency is increased.
Description
Technical field:
The present invention relates to a kind of compression coding method for signal source of wireless communication field, mainly relate to a kind of utmost point Video Compression at Low Bit-rate method based on wavelet transformation, the method mainly is the effective transmission that is applicable to carry out utmost point low bit-rate video information under the anti-interference environment of satellite, belongs to wireless communication technology field.
Background technology:
Exist various interference and noise in the wireless channel, the satellite channel environment is more abominable, impact for vision signal is very big, and transmission video signal requires encoding scheme to have very high compression ratio in wireless channel, how to utilize wireless channel to transmit the emphasis that vision signal becomes people's research.In satellite communication system, for tackling extremely strong electromagnetic interference, utmost point Low Rate Code Technology can guarantee under the prerequisite of certain mass so that the speed of communication is low as much as possible, thereby saves bandwidth, raises the efficiency, and guarantees minimum essential communication.
H.263, H.264(MPEG4 standard in the low bit rate video coding method mainly is) and the Chinese Industrial Standards (CIS) AVS (Chen Kangli that studying, Wang Shaolin. the compression of moving image under the extremely low code check. Chinese image graphics journal, Vol.5 (A), No.7, July2000.).Uniting of several technology makes to finish often for standard.Very H.263 the compressed encoding of low bit rate mainly comprises the main contents with the MPEG-4 standard.At present most widely used communication channel is low bit rate channel (very low bit rate≤64Kbits/s) in the reality, channel width such as PSTN is 64Kbits/s(Su Jie, Cao Zhongsheng, Feng Yucai. Very low bit-rate video coding [J]. Journal of Computer Research and Development, 1998, (04) .), and the bandwidth of wireless channel often only has several Kbit/s.Nearest H.263, the MPEG-4 coding standard is exactly the coding standard that is adapted to low code check.H.263 be the ITU-T standard that is used for the earliest Low Bit-Rate Video Coding, it be the H.261 standard formulated take ITU-T as the basis, take hybrid coding as core technology, organize code stream according to the form of layering.Wherein, H.263 adopted the method for bilinear interpolation to realize the estimation of half-pixel accuracy.It has the unrestricted motion arrow pattern, based on the arithmetic coding pattern of syntax, and advanced prediction mode, four kinds of the encoding options of consulting to add such as PB frame pattern, thus further improved code efficiency.(Wang Daming, Huang Huiqun, Guo Jianxin. the research of PB frame technique in the Very Low Bit Rates Video Coding. signal is processed, Vol.19, No.3,2003.) but it has still kept traditional dct transform mode of carrying out interframe encode and inter prediction encoding, therefore still has the vestige of piece when extremely hanging down code check.H.263 mainly adopt the technology such as block motion compensation discrete cosine transform, hybrid coding, plurality of optional pattern.In low code check system, the common amplitude of image motion is less, and mode is simple.A kind of based on the improving one's methods of adaptive judgement, the improvement algorithm on standard H.263.MPEG-4 has larger promotion to Low Bit-rate Coding, code check has wide region adaptability (5k-10Mbit/s), emphasized the coding standard of low code check (5-64kb/s), simultaneously the content of standard has H.263 been carried out compatible extensions, before inheriting, also relate to the technology such as wavelet transformation, motion compensation on the basis of standard technique.
Based on the Video coding of 3 D wavelet transformation be the two-dimensional space Wavelet Image Compression to the simple popularization in 3 D video space (Hu Yi and, Yu Sile. based on the very low bitrate video coding technology of wavelet transformation. TV tech, NO.210, Dec.1999.).The video compression algorithm based on wavelet transformation that occurs at present mainly can be divided into following four kinds dissimilar:
(1) contains the 3 D wavelet video coding (MC-3DMC) of motion compensation
(2) based on the small wave video coding (DWT-MC) of transform domain motion compensation
(3) based on the small wave video coding (MC-DWT) of spatial domain motion compensation
(4) based on the embedded video of 3 D wavelet transformation coding (3D-DWT)
Wavelet transformation has both direction at first: a kind of is that wavelet transformation combines with motion compensation, can divide several situations again in this case: can utilize first motion compensation to eliminate time redundancy, then utilize wavelet transformation to eliminate spatial redundancy; Also can utilize first wavelet transformation to eliminate spatial redundancy, again transform frame be carried out motion compensation and eliminate time redundancy.Another direction is that time and spatial domain are all carried out wavelet transformation, in view of the scalability of wavelet transformation, is suitable for the situation that gradual transmission of video and the network bandwidth have fluctuation.
Shapiro proposed EZW(Embedded Zerotree Wavelet in 1993) algorithm, remainder organizational form (the embedded image coding .IEEE signal processing transactions of Shapiro J.M.. use zero-tree wavelet coefficient of wavelet coefficient have been opened, 1993,41 (12): 3445-3462.), 1996, Said and Pearlman improve remainder encryption algorithm, SPIHT(Set Partitioning in Hierarchical Tree has been proposed) algorithm, it can more effectively organize wavelet coefficient (A.Said and W.A.Pearlman. Circuits and Systems transactions based on the quick and effective method for encoding images .IEEE video technique of the multistage tree set partitioning of small echo, vol.6, pp.43-250, June1996.).1997, Kim and Pearlman were generalized to Video coding with the SPIHT method, had proposed three-dimensional multistage tree set partitioning method (3D-SPIHT).Calendar year 2001, Khan etc. with two dimensional image virtual tree method utilization and extention to three dimensions, the 3D-VSPIHT coding method has been proposed, this method is than conventional method higher (E.Khanand on efficient, M.Ghanbari. use the very low bit rate image compression encoding of the multistage tree set partitioning of virtual small echo, IEEE electronics wall bulletin, 2001,37 (1): 40-42.).Document " the three-dimensional set partitioning in hierarchical trees method for video coding of improved wavelet field ". (Yang Chunling, Yu Yinglin. the communication journal, 200l, 22 (5): 86-92.), " redundant optimization three-dimensional VSP IHT method for video coding between fully reducing to set. " (Ding Wenqi, Hu Jia, Zhang Liming. computer-aided design and graphics journal, 2005,17 (3): 563-569.) carried out corresponding optimization and improvement on the basis of above-mentioned algorithm.
Summary of the invention:
1, purpose: in order to realize that the video multimedia data are in the high efficient and reliable transmission that is subject on the satellite channel of external interference, utmost point Video Compression at Low Bit-rate becomes the emphasis of research, in recent years study hotspot based on the Study on Video Coding Algorithm of small echo, replace in the existing video standard of video take the kernel kernal mapping method of dct transform as the basis, dct transform blocking effect easily occurs under low code check, and wavelet transformation has the characteristics of concentration of energy and do not produce blocking effect, simultaneously can also utilize the similitude of its equidirectional subband to carry out subsequent treatment, be conducive to improve compression ratio, thereby improve code efficiency.Traditional 3 D video compression algorithm based on small echo is followed two dimensional image next code method, the wavelet coefficient tissue is not carried out further optimization process.The purpose of this invention is to provide a kind of utmost point Video Compression at Low Bit-rate method based on wavelet transformation, it inherits and has improved traditional video small echo 3D-SPIHT algorithm, utilize virtual tree to organize wavelet conversion coefficient with the form that the coefficient blocking organization combines, simultaneously the wavelet coefficient of different sub-band carried out certain differentiation according to the difference of threshold value, adopt many threshold values sub-band coding optimization method, the significant coefficient priority encoding, reach the purpose that represents more effective information with bit number still less, compression ratio can be improved, thereby compression efficiency can be improved.
2, technical scheme: principal character of the present invention is: at first vision signal is carried out 3 D wavelet transformation, obtain wavelet conversion coefficient, next step main task is that wavelet coefficient is carried out follow-up tissue arrangement, utilizes virtual tree method and the wavelet coefficient method of partition implemented in Image Coding to carry out combination.The virtual tree method is that the lowest frequency coefficient LLN to highest decomposition stage carries out virtual decomposition, in fact do not carry out actual wavelet decomposition, but gather like that division according to wavelet decomposition, after virtual by one-level, one tree can be organized more wavelet coefficient, thereby can improve code efficiency under certain condition, do not carrying out before the virtual decomposition, the corresponding one tree of node of the wavelet coefficient that level Four is decomposed, and carry out virtual after, the node that virtual one-level is decomposed can comprise 4 nodes that original level Four is decomposed wavelet coefficient.The wavelet coefficient method of partition is that the coefficient after the decomposition (comprising virtual decomposition) is divided with the form of piece, four wavelet coefficients are divided into a piece, be equivalent to a node, in follow up scan, it is used as a node and processes, determine whether the coefficient in the piece is processed separately according to the result who processes.Simultaneously because the wavelet coefficient importance of different sub-band is different, traditional SPIHT method is not distinguished different sub-band when coding, according to statistical law, more past low frequency often, the amplitude of coefficient is larger, corresponding also more important, the present invention also utilizes these characteristics, carries out many threshold optimizations and processes.
Fig. 1 has provided the system block diagram that adopts the inventive method.
System's 11 parts are at first carried out 3D-DWT to video sequence, comprise that namely the 3 D wavelet in time and space decomposes.Order on time decomposition and spatial decomposition does not affect the wavelet coefficient that finally obtains, and decomposes so can carry out first small echo time one dimension, then time low frequency and the high frequency that obtains is carried out spatial decomposition.Time dimension decomposes: because the time dimension limit in order to be easy to BORDER PROCESSING, utilizes the Haar wavelet basis that time dimension is carried out wavelet decomposition.The space dimension is decomposed: the space dimension is adopted the Daubechies9/7 wavelet basis, spatial domain is carried out wavelet decomposition.
12 parts space lowest frequency to the time lowest frequency on the basis that 11 parts are decomposed is substantially carried out virtual decomposition, the time portion of this moment can not be carried out further virtual decomposition generally speaking, if image sets (GOP) number of image frames of selecting enough can satisfy greatly further virtual decomposition, but tend to increase complexity and the processing time of processing, so generally in the experiment select 8 frames or 16 frames carry out wavelet transformation as one group.The wavelet coefficient of 13 parts after to actual and virtual decomposition carries out further piecemeal processing, take four wavelet coefficients as one group.14 parts are many threshold optimizations, carry out threshold value statistics and next code processing according to the coefficient amplitude of different sub-band.15 parts be in the situation that above work all conduct wavelet coefficient is scanned and encodes.The 16th, the inverse process of coding.17 is the 3 D wavelet inverse transformation, finally obtains the video sequence of reconstruct.
Fig. 2 has provided the realization schematic diagram that virtual tree is divided among the present invention.
When wavelet coefficient is organized, use for reference the concept of two dimensional image virtual tree, the decomposition coefficient that the space is tieed up carries out virtual decomposition, and time dimension is because frame number limits, and general low frequency does not carry out virtual decomposition.
If picture frame group (GOF) is 8 frames, CIF image size is 352 * 288, carries out that the size of LLL3 is 44 * 36 * 1 after three grades of decomposition.It is carried out secondary virtual, virtual mode is consistent with the two-dimensional process mode.
How defining wavelet tree is the key that wavelet coefficient is processed, and also is the emphasis of follow-up SPIHT coding, and the size of establishing ROOT is W
n* H
n, according to the difference of the residing subband of coefficient, define three kinds of dissimilar children relations:
(1) if (i, j, k) under ROOT because the node number (such as 11 * 9) under the ROOT can not be according to four groupings, then node correspondence is with a node of the different directions same position of decomposition level, then defines its children and is
Have seven offsprings, three offsprings are arranged in the time lowest band, four lowest spatial frequency bands that are positioned at high frequency at the same level.
(2) if (i, j, k) not at ROOT, still also is in virtual part, then it has four offsprings { (2i, 2j, k), (2i+1,2j, k), (2i, 2j+1, k), (2i+1,2j+1, k) }.
(3) if (i, j, k) falls non-virtual subnet band portion, eight offsprings are arranged on the traditional sense, be respectively
The wavelet coefficient of HFS is relatively inessential, for the more effective wavelet coefficient of organizing HFS, if three grades of true wavelet coefficients that decompose are divided into groups with 2 * 2, namely from LHLA, LLHA, LHHA, HHLA, HLHA and the HHHA of virtual decomposition, the set of wavelet coefficients that wavelet coefficient is corresponding four 2 * 2, LHL3, LLH3, LHH3, HHL3, HLH3, HHH3 and followingly all adopt this mode to organize wavelet coefficient.
Fig. 3 is the schematic diagram of block coefficient tissue among the present invention.
Wavelet coefficient tissue to image carries out three-dimensional extension, 2 * 2 coefficient block of two dimension are expanded to three-dimensional, form 2 * 2 * 2 coefficient block, the node of coefficient block has one 2 * 2 * 2 coefficient block as children in subband equidirectional with it, the wavelet coefficient of minimum decomposition level does not have children, low frequency sub-band in the highest decomposition stage is as tree root, and the direct offspring of each tree root is 7 nodes except itself that are in identical decomposition level with it.Root node is comprised of 8 coefficients, and except a coefficient infertility, all the other 7 coefficients point to 2 * 2 * 2 coefficient block of the similar position of identical decomposition level.
As shown in Figure 3, the position coordinates of (x, y, z) expression 3 D wavelet coefficient, H represents the set of all tree roots, it comprises all nodes of the highest decomposition stage of wavelet decomposition, if the frame number of a GOF is F, after three grades of wavelet decomposition, lowest frequency is of a size of W
n, L
n, F
nDistinguish wide, long and frame number, the direct offspring of root is
Except leaf node was had no children, the children of intermediate node were:
If establish C
X, y, zThe wavelet coefficient values that expression (x, y, z) is located then has
Judge the importance of wavelet coefficient, if 1 expression wavelet coefficient is important for current threshold value, otherwise inessential, T is set of expression,
Concrete process is consistent with SPIHT.
To sum up, a kind of utmost point Video Compression at Low Bit-rate method based on wavelet transformation of the present invention, it mainly is to be subject to carrying out MIN communication under the external interference condition in order to adapt to satellite channel, reaches the purpose of extremely low code check transmission, and the implementation step of the method is as follows:
Decoding end and coding side process contrary, at coding side:
Step 1: at first the GOF image sets is carried out 3 D wavelet and decompose, time and spatial decomposition are selected suitable decomposed class according to the size of image sets and the form of video (CIF, QCIF, SIF etc.), obtain the coefficient matrix of wavelet decomposition after the decomposition.Now with GOF=8, CIF(352 * 288) the Foreman video of form is that example is carried out three grades of times and spatial decomposition, and each sub-band coefficients matrix size is as shown in table 1 after decomposing.
Each sub-band coefficients matrix size after table 1 wavelet decomposition
| Subband | LLL3 | LHL3 | LLH3 | LHH3 |
| The |
44×36×1 | 44×36×1 | 44×36×1 | 44×36×1 |
| Subband | ----- | LHL2 | LLH2 | LHH2 |
| The coefficient matrix size | ----- | 88×72×2 | 88×72×2 | 88×72×2 |
| Subband | ------ | LHL1 | LLH1 | LHH1 |
| The coefficient matrix size | ------ | 176×144×4 | 176×144×4 | 176×144×4 |
Each sub-band coefficients maximum of lowest frequency analysis frames after table 2 wavelet decomposition
| Subband | LLL3 | LHL3 | LLH3 | LHH3 | LHL2 |
| Coefficient value | 2679.7 | 404.5860 | 629.6811 | 289.2501 | 282.4217 |
| Subband | LLH2 | LHH2 | LHL1 | LLH1 | LHH1 |
| Coefficient value | 366.2796 | 148.8928 | 290.9120 | 144.8454 | 60.5448 |
Foreman is carried out after 3 D wavelet decomposes; the wavelet coefficient maximum of each subband of time low frequency sub-band is as shown in table 2; can find out from highest decomposition stage and reduce gradually to minimum decomposition level wavelet coefficient amplitude; utilize this point when wavelet coefficient is encoded, to carry out certain processing to important subband, to reach the purpose of protection significant coefficient.
Step 2: divide tree minute block operations to coefficient matrix.At first coefficient matrix is carried out one-level or the virtual decomposition of secondary, then the coefficient after the virtual decomposition (can be used as normal decomposition treats) is carried out piecemeal and process.
Fig. 3 is decomposed into example take GOF as three grades of 8 time and spaces, the LLLt subband divides into groups with 2 * 2 to lowest frequency subband LL3, there is not child at air-frame No. 1 of every group, but be equipped with a child in the LLHt identical bits, remaining is 2,3 years old, No. 4 respectively at air-frame with decomposition level HL3, LH3, the equidirectional of HH3 have four children, in the LLHt same position child are arranged also.
Each coefficient of HL3, LH3 and HH3 is the equidirectional coefficient block children that have 2 * 2 of next decomposition level.The situation of other frames is similar, and for the LL3 subband, except LLLt and Ht, 1,2,3, No. 4 coefficient all is equipped with two offsprings (only drawn among the figure 1 and 4,2 and 3 similar) in the identical bits of next decomposition level time frame of time dimension.
If (i, j, k) is No. 1 coefficient, then its children are (i, j, k+1) or (i, j, k+1), (i, j, k+2).
Step 3: SPIHT cataloged procedure
Traditional SPIHT coding step is as follows:
In concrete algorithm, the diversity rule definition two types set: the set D and the set L.The D type
Table represents all child nodes, and the L-type table represents non-child descendants node.
Spiht algorithm is with the intermediate treatment of three chained lists realizations to wavelet coefficient, i.e. LSP(List of Significant Pixels), LIP(List of Insignificant Pixels) and LIS(List of Insignificant Sets).
The SPIHT cataloged procedure is minute four steps specifically:
The first step: initialization.The initialization codes threshold value is got
C wherein
I, jWavelet coefficient threshold value T=2 then for (i, j) point
nTo the sequential chained list initialization, LSP is made as empty set, and LIP is initialized as the wavelet coefficient coordinate (whole coefficients of the 3rd layer of decomposition) of highest decomposition, and LIS is initialized as descendants's wavelet coefficient coordinate among the LIP, initially be made as the D type.The order of coordinate is the same with ZT coding (EZW) among LIS and the LIP.
Second step: ordering scanning.
At first LIP is scanned, each node among the LIP is carried out importance judge, if then important greater than threshold value, output 1, otherwise export 0, if 1 is exported the symbol of this point and it is moved on to the LSP afterbody, delete from LIP.
Next carries out test of significance to the set among the LIS and outputs test result, if each wavelet coefficient of set thinks then that less than the coding threshold value this set is inessential, output 0.Importantly then export 1.Under the material circumstance,
If type is D type (comprising child and non-child nodes), judge the importance of each list item, if important output code flow 1.Wherein child nodes is scanned, if important in code stream output 1 and output symbol, it is moved to the LSP afterbody; If inessential then to code stream output 0, and add it to LIP afterbody.Judge whether non-child's set is empty, and non-NULL then moves to it LIS afterbody, is labeled as the L list item.If sky is then deleted D type list item from LIP.
If it is L and important that the list item type is arranged among the LIS, export 1 to code stream, and L is split into four D type list items, add LIS table tail to, and the current set of deletion from LIS.
The 3rd step: refinement
Wavelet coefficient corresponding to coefficient coordinate that the higher level is scanned in the LSP table that obtains is converted to binary system, exports N bit of each coefficient to meticulous code stream.
The 4th step: upgrade
N is subtracted 1, at current threshold value T=2
N-1Under re-start second step scanning.
In the methods of the invention, will carry out certain improvement to the SPIHT coding, detailed process is as follows:
Each coefficient of HL3, LH3 and HH3 is the equidirectional coefficient block children that have 2 * 2 of next decomposition level.The situation of other frames is similar, and for the LL3 subband, except LLLt and Ht, 1,2,3, No. 4 coefficient all is equipped with two offsprings (only drawn among the figure 1 and 4,2 and 3 similar) in the identical bits of next decomposition level time frame of time dimension.
If (i, j, k) is No. 1 coefficient, then its children are (i, j, k+1) or (i, j, k+1), (i, j, k+2).
Remaining and image coefficient are organized similar, are beneficial to like this two dimension and the three-dimensional form that combines.
Because the similitude of adjacent wavelet coefficient is beneficial to and utilizes piecemeal further to organize wavelet coefficient.Adjacent several wavelet coefficients are organized into a piece as a node, can be that laterally two adjacent wavelet coefficients are bound one, or vertically two adjacent coefficients are bound, can also be four wavelet coefficients of a square, also can be four rectangular wavelet coefficients, the like, but wavelet coefficient is more, complexity is higher, can only consider the situation of two or four.
Take four pieces as one group as example, remaining organizational form is equivalent to exactly 1,2,3,4 and represents respectively one 2 * 2 coefficient block with top consistent.
Concrete minute three set, LSP={}; LIP={ (i, j, k) (i, j, k) ∈ H, H is tree root }; LIS={ (i, j, k) D (i, j, k) ∈ H, and the descendants is arranged }.
Can be by following initialization:
LSP＝{}；
LIP＝{（1，1，1），（1，3，1），（3，1，1），（3，3，1）}；
LIS＝{（1，3，1）D，（3，1，1）D，（3，3，3）D}。
The next code step is consistent with traditional SPIHT mode, and the processing mode of some node need to be treated especially, and concrete step will be in follow-up introduction.Can simplify wavelet tree like this, reduce complexity, organize efficiently again wavelet coefficient simultaneously.If Wavelet Coefficient Blocks is inessential, then can represent four original wavelet coefficients with 1 bit.
The next code step is as follows:
Step 1: initialization threshold value and ordered list, ordered list is by top initialization, threshold value
Step 2: ordering scanning
Step 2.1: the coefficient block among the sequential scanning LIP (i, j, k)
If coefficient block is important, namely
Output 1 and respectively each coefficient of this piece is carried out test of significance, if important coefficient then exports 1 and sign bit, and this coefficient moved to LSP, if inessentially then export 0, if four coefficients of a coefficient block are all handled then it is deleted from LIP.
If this coefficient block is inessential, output 0.
Step 2.2: the list item among the processed in sequence LIS (i, j, k) D/L processes, and D type list item is carried out respectively different processing with the L-type list item.
Step 2.2.1: if D type list item is judged importance, if importantly then export 1, and D type list item is resolved into L-type list item and four coefficient sub-blocks, then four coefficient sub-blocks are processed successively:
If the coefficient sub-block is important, then export 1, and four coefficients of coefficient sub-block are processed, if the important output 1 of coefficient and sign bit, and it is added into the LSP afterbody, then export 0 if coefficient is inessential.
If coefficient block is inessential, then export 0.
To the L-type list item that D type table decomposes, judge whether it is empty set, if empty explanation (i, j, k) only has child nodes, it is deleted from LIS; If not empty, then it is moved to LIS table afterbody as the L list item.
If D type list item is inessential, output 0.
Step 2.2.2: if the L-type list item is judged its importance, if important output 1, and it is decomposed into four D type list items, and add the LIS afterbody to, the L-type table is deleted from LIS; If inessential, then export 0.
Step 3: fine scanning
To the coefficient among the LSP (i, j, k), if not add in the step 2 of just having carried out, then output | C
I, j, k| binary representation in N important bit, T=2
NBe threshold value.
Step 4: threshold value is upgraded
With T=T/2, scanning and fine scanning sort next time.
3, advantage and effect:
This method that the present invention proposes is to be subject to carrying out MIN communication under the external interference condition in order to adapt to satellite channel, reaches the purpose of extremely low code rate information transmission.Utilize the thought of virtual tree, the wavelet transformation subband has been carried out further division, be conducive to organize more wavelet coefficient, utilize simultaneously the thought of piecemeal, also be based on this purpose, if wavelet coefficient is important, then processing mode is consistent with traditional approach, if the wavelet coefficient of a node is inessential, then can represent the information that original several bits of needs represent with a bit, thereby can utilize less bit to represent useful information, reduce number of coded bits, reach the requirement of extremely low code check.
Concrete advantage overview is as follows:
(1) utilizes the mode of virtual tree and section thinking combination, virtual tree is in order to prolong wavelet tree, thereby can organize more wavelet coefficient on the one tree, and section thinking is to utilize a plurality of coefficients to carry out scan process as a node, also be for the integrated tissue wavelet coefficient, final purpose all is to organize more wavelet coefficient in order to reach with bit number still less, if coefficient is inessential, then can represent one tree or a node with a bit, organize efficiently wavelet coefficient thereby reach, same information can represent with less bit, thereby reduces code check.
(2) traditional SPIHT method is not distinguished each decomposition subband, add up from amplitude, more the coefficient of low frequency is often larger, relatively also just important, and the coefficient amplitude of high-frequency sub-band is little, importance is taken second place, invention has utilized many threshold optimizations technology, the maximum amplitude of each subband is added up in compromise, finds the coding threshold value of each subband, carries out many threshold optimizations codings.
Description of drawings:
Fig. 1 has provided the system block diagram that adopts the inventive method.
Fig. 2 has provided the realization schematic diagram that virtual tree is divided among the present invention.
Fig. 3 is the realization schematic diagram of block coefficient organizational form among the present invention.
Symbol description is as follows among the figure:
11 is the 3 D wavelet decomposing module, 12 is virtual minute tree module, 13 is wavelet coefficient piecemeal module, 14 is many threshold optimizations part, 15 is traditional SPIHT coded portion, wherein 12,13,14,15 form M3D-SPIHT(Modified 3D-SPIHT) coded portion, 16 is the M3D-SPIHT decoded portion, 17 is 3 D wavelet inverse transformation part.
LLLt, LLHt, LHt, Ht represent low frequency and the HFS of time decomposition, LLL, LLH, LH, H also represent low frequency and the HFS of time decomposition direction.
LLN, LHN, HLN, HHN represent the level of spatial decomposition, each subband of vertical and diagonal.
Embodiment:
Implement such as Fig. 1, Fig. 2, Fig. 3, take 16 frames as a GOF, the CIF take 352 * 288 is test pattern, frame per second is made as 10fps, is fit to the requirement of extremely low code check transmission.Utilize the Haar small echo to decompose to time dimension, space utilization db9/7 biorthogonal wavelet carries out two dimension and decomposes.
Concrete steps are as follows:
Step 1: at first, image is carried out 3D-DWT.Specifically utilize first the Haar wavelet basis that time orientation is carried out three grades of wavelet decomposition, then utilize the Daubechies9/7 small echo that each chronon band after decomposing is carried out three grades of spatial decomposition, the LLL3 size that obtains is 44 * 36 * 2, again lowest frequency LLL3 is carried out the virtual decomposition of one-level, the LLLA size that obtains after the virtual decomposition is 22 * 18 * 1.
Step 2: secondly, the coefficient that decomposes is carried out piecemeal process, the coefficient block with 2 * 2 is as a node, and follow-up diversity is consistent with traditional SPIHT.
Step 3: SPIHT coding
Concrete minute three set, LSP={}; LIP={ (i, j, k) (i, j, k) ∈ H, H is tree root };
LIS={ (i, j, k) D (i, j, k) ∈ H, and the descendants is arranged }.
Can be by following initialization:
LSP={}；
LIP＝{（1，1，1），（1，3，1），（3，1，1），（3，3，1）}；
LIS＝{（1，3，1）D，（3，1，1）D，（3，3，3）D}。
The next code step is as follows:
Step 1: initialization threshold value and ordered list, ordered list is by top initialization, threshold value
Step 2: ordering scanning
Step 2.1: the coefficient block among the sequential scanning LIP (i, j, k)
If coefficient block is important, namely
Output 1 and respectively each coefficient of this piece is carried out test of significance, if important coefficient then exports 1 and sign bit, and this coefficient moved to LSP, if inessentially then export 0, if four coefficients of a coefficient block are all handled then it is deleted from LIP.This is where to move time of what situation, but this be a few days ago thing also really bad resolution you
If this coefficient block is inessential, output 0.
Step 2.2: the list item among the processed in sequence LIS (i, j, k) D/L processes, and D type list item is carried out respectively different processing with the L-type list item.
Step 2.2.1: if D type list item is judged importance, if importantly then export 1, and D type list item is resolved into L-type list item and four coefficient sub-blocks, then four coefficient sub-blocks are processed successively:
If the coefficient sub-block is important, then export 1, and four coefficients of coefficient sub-block are processed, if the important output 1 of coefficient and sign bit, and it is added into the LSP afterbody, then export 0 if coefficient is inessential.
If coefficient block is inessential, then export 0.
To the L-type list item that D type table decomposes, judge whether it is empty set, if empty explanation (i, j, k) only has child nodes, it is deleted from LIS; If not empty, then it is moved to LIS table afterbody as the L list item.
If D type list item is inessential, output 0.
Step 2.2.2: if the L-type list item is judged its importance, if important output 1, and it is decomposed into four D type list items, and add the LIS afterbody to, the L-type table is deleted from LIS; If inessential, then export 0.
Step 3: fine scanning
To the coefficient among the LSP (i, j, k), if not add in the step 2 of just having carried out, then output | C
I, j, k| binary representation in N important bit, T=2
NBe threshold value.
Step 4: threshold value is upgraded
With T=T/2, scanning and fine scanning sort next time.
Claims (1)
1. utmost point Video Compression at Low Bit-rate method based on wavelet transformation, it is characterized in that: the implementation step of the method is as follows:
Decoding end and coding side process contrary, at coding side:
Step 1: at first the GOF image sets is carried out 3 D wavelet and decompose, time and spatial decomposition are selected suitable decomposed class according to the size of image sets and form CIF, QCIF, the SIF of video, obtain the coefficient matrix of wavelet decomposition after the decomposition; Now with GOF=8, CIF(352 * 288) the Foreman video of form is that example is carried out three grades of times and spatial decomposition, and each sub-band coefficients matrix size is as shown in table 1 after decomposing
Each sub-band coefficients matrix size after table 1 wavelet decomposition
Each sub-band coefficients maximum of lowest frequency analysis frames after table 2 wavelet decomposition
Foreman is carried out after 3 D wavelet decomposes, the wavelet coefficient maximum of each subband of time low frequency sub-band is as shown in table 2, find out thus from highest decomposition stage and reduce gradually to minimum decomposition level wavelet coefficient amplitude, utilize this point when wavelet coefficient is encoded, important subband to be processed, to reach the purpose of protection significant coefficient;
Step 2: divide tree minute block operations to coefficient matrix; At first coefficient matrix is carried out one-level or the virtual decomposition of secondary, then the coefficient after the virtual decomposition is carried out piecemeal and process;
Be decomposed into example take GOF as three grades of 8 time and spaces, the LLLt subband divides into groups with 2 * 2 to lowest frequency subband LL3, there is not child at air-frame No. 1 of every group, but be equipped with a child in the LLHt identical bits, remaining is 2,3 years old, No. 4 respectively at air-frame with decomposition level HL3, LH3, the equidirectional of HH3 have four children, in the LLHt same position child are arranged also;
Each coefficient of HL3, LH3 and HH3 is the equidirectional coefficient block children that have 2 * 2 of next decomposition level, the situation of other frames is similar, for the LL3 subband, except LLLt and Ht, 1,2,3, No. 4 coefficient all is equipped with two offsprings in the identical bits of next decomposition level time frame of time dimension;
If (i, j, k) is No. 1 coefficient, then its children are (i, j, k+1) or (i, j, k+1), (i, j, k+2);
Step 3: the SPIHT cataloged procedure, its specific implementation process is as follows:
Each coefficient of HL3, LH3 and HH3 is the equidirectional coefficient block children that have 2 * 2 of next decomposition level, the situation of other frames is similar, for the LL3 subband, except LLLt and Ht, 1,2,3, No. 4 coefficient all is equipped with two offsprings in the identical bits of next decomposition level time frame of time dimension; If (i, j, k) is No. 1 coefficient, then its children are (i, j, k+1) or (i, j, k+1), (i, j, k+2);
Remaining and image coefficient are organized similar, are beneficial to like this two dimension and the three-dimensional form that combines;
Because the similitude of adjacent wavelet coefficient is beneficial to and utilizes piecemeal further to organize wavelet coefficient; Adjacent several wavelet coefficients are organized into a piece as a node, can be that laterally two adjacent wavelet coefficients are bound one, or vertically two adjacent coefficients are bound, can also be four wavelet coefficients of a square, also can be four rectangular wavelet coefficients, the like, but wavelet coefficient is more, complexity is higher, only considers the situation of two or four;
Take four pieces as one group as example, remaining organizational form is equivalent to exactly 1,2,3,4 and represents respectively one 2 * 2 coefficient block with top consistent;
Concrete minute three set, LSP={}; LIP={ (i, j, k) (i, j, k) ∈ H, H is tree root }; LIS={ (i, j, k) D (i, j, k) ∈ H, and the descendants is arranged };
Press following initialization:
LSP＝{}；
LIP＝{（1，1，1），（1，3，1），（3，1，1），（3，3，1）}；
LIS＝{（1，3，1）D，（3，1，1）D，（3，3，3）D}；
The next code step is consistent with traditional SPIHT mode, and the processing mode of some node need to be treated especially, if Wavelet Coefficient Blocks is inessential, then represents four original wavelet coefficients with 1 bit;
The next code step is as follows:
Step 1: initialization threshold value and ordered list, ordered list is by top initialization, threshold value
Step 2: ordering scanning
Step 2.1: the coefficient block among the sequential scanning LIP (i, j, k);
If coefficient block is important, namely
Output 1 and respectively each coefficient of this piece is carried out test of significance, if important coefficient then exports 1 and sign bit, and this coefficient moved to LSP, if inessentially then export 0, if four coefficients of a coefficient block are all handled then it is deleted from LIP;
If this coefficient block is inessential, output 0;
Step 2.2: the list item among the processed in sequence LIS (i, j, k) D/L processes, and D type list item is carried out respectively different processing with the L-type list item;
Step 2.2.1: if D type list item is judged importance, if importantly then export 1, and D type list item is resolved into L-type list item and four coefficient sub-blocks, then four coefficient sub-blocks are processed successively:
If the coefficient sub-block is important, then export 1, and four coefficients of coefficient sub-block are processed, if the important output 1 of coefficient and sign bit, and it is added into the LSP afterbody, then export 0 if coefficient is inessential;
If coefficient block is inessential, then export 0;
To the L-type list item that D type table decomposes, judge whether it is empty set, if empty explanation (i, j, k) only has child nodes, it is deleted from LIS; If not empty, then it is moved to LIS table afterbody as the L list item;
If D type list item is inessential, output 0;
Step 2.2.2: if the L-type list item is judged its importance, if important output 1, and it is decomposed into four D type list items, and add the LIS afterbody to, the L-type table is deleted from LIS; If inessential, then export 0;
Step 3: fine scanning
To the coefficient among the LSP (i, j, k), if not add in the step 2 of just having carried out, then output | C
I, j, k| binary representation in N important bit, T=2
NBe threshold value;
Step 4: threshold value is upgraded
With T=T/2, scanning and fine scanning sort next time.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2013100196008A CN103024399A (en) | 2013-01-18 | 2013-01-18 | Wavelet transform based extreme-low bit-rate video compressing and coding method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2013100196008A CN103024399A (en) | 2013-01-18 | 2013-01-18 | Wavelet transform based extreme-low bit-rate video compressing and coding method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103024399A true CN103024399A (en) | 2013-04-03 |
Family
ID=47972471
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2013100196008A Pending CN103024399A (en) | 2013-01-18 | 2013-01-18 | Wavelet transform based extreme-low bit-rate video compressing and coding method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103024399A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103647959A (en) * | 2013-12-23 | 2014-03-19 | 国网安徽省电力公司信息通信分公司 | Method for transmitting high-definition digital photographs in very low bit rate application |
| WO2017011943A1 (en) * | 2015-07-17 | 2017-01-26 | 华为技术有限公司 | Signal transmission method, apparatus, and device |
| CN107257473A (en) * | 2017-07-18 | 2017-10-17 | 西安电子科技大学 | A kind of efficient image compression algorithm |
| CN109035350A (en) * | 2018-08-02 | 2018-12-18 | 辽宁师范大学 | Improved SPIHIT image coding and decoding method based on energy leakage and amplification |
| CN112995637A (en) * | 2021-03-10 | 2021-06-18 | 湘潭大学 | Multi-section medical image compression method based on three-dimensional discrete wavelet transform |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020181593A1 (en) * | 2001-05-29 | 2002-12-05 | Tinku Acharya | Method and apparatus for three-dimensional wavelet transform |
| CN1564603A (en) * | 2004-04-01 | 2005-01-12 | 复旦大学 | Organization method of 3-D virtual SPIHT tree having time limited |
| CN1564604A (en) * | 2004-04-08 | 2005-01-12 | 复旦大学 | Gradation tree set partitioning image compression method based on tree shaped structure |
-
2013
- 2013-01-18 CN CN2013100196008A patent/CN103024399A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020181593A1 (en) * | 2001-05-29 | 2002-12-05 | Tinku Acharya | Method and apparatus for three-dimensional wavelet transform |
| CN1564603A (en) * | 2004-04-01 | 2005-01-12 | 复旦大学 | Organization method of 3-D virtual SPIHT tree having time limited |
| CN1564604A (en) * | 2004-04-08 | 2005-01-12 | 复旦大学 | Gradation tree set partitioning image compression method based on tree shaped structure |
Non-Patent Citations (5)
| Title |
|---|
| 张勇东等: "高性能三维小波视频编码方法", 《通信技术》 * |
| 朱华亮等: "基于系数块的改进SPIHT 算法", 《计算机工程》 * |
| 李跃明: "一种基于改进SPIHT的图像压缩算法研究", 《微计算机信息》 * |
| 杨春玲等: "基于三维小波变换嵌入式视频压缩算法的研究", 《电子学报》 * |
| 杨春玲等: "改进的小波域三维等级树集分割视频编码方法", 《通信学报》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103647959A (en) * | 2013-12-23 | 2014-03-19 | 国网安徽省电力公司信息通信分公司 | Method for transmitting high-definition digital photographs in very low bit rate application |
| WO2017011943A1 (en) * | 2015-07-17 | 2017-01-26 | 华为技术有限公司 | Signal transmission method, apparatus, and device |
| CN106797481A (en) * | 2015-07-17 | 2017-05-31 | 华为技术有限公司 | Method for transmitting signals, device and equipment |
| CN107257473A (en) * | 2017-07-18 | 2017-10-17 | 西安电子科技大学 | A kind of efficient image compression algorithm |
| CN107257473B (en) * | 2017-07-18 | 2020-01-10 | 西安电子科技大学 | Efficient image compression algorithm |
| CN109035350A (en) * | 2018-08-02 | 2018-12-18 | 辽宁师范大学 | Improved SPIHIT image coding and decoding method based on energy leakage and amplification |
| CN112995637A (en) * | 2021-03-10 | 2021-06-18 | 湘潭大学 | Multi-section medical image compression method based on three-dimensional discrete wavelet transform |
| CN112995637B (en) * | 2021-03-10 | 2023-02-28 | 湘潭大学 | Multi-section medical image compression method based on three-dimensional discrete wavelet transform |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103024399A (en) | Wavelet transform based extreme-low bit-rate video compressing and coding method | |
| CN1255218A (en) | Apparatus and method for encoding wavelet trees generated by wavelet-based coding method | |
| CN103581691B (en) | A kind of towards sparse coefficient efficiently can parallel image coding method | |
| CN1650634A (en) | Scalable wavelet based coding using motion compensated temporal filtering based on multiple reference frames | |
| Fang et al. | Interpolation-based direction-adaptive lifting DWT and modified SPIHT for image compression in multimedia communications | |
| CN1669326A (en) | Wavelet based coding using motion compensated filtering based on both single and multiple reference frames | |
| CN111432210A (en) | Point cloud attribute compression method based on filling | |
| CN101848390B (en) | Image compression secure coding method based on multidirectional sparse representation | |
| CN101132529A (en) | Wavelet image compression method based on fast and successive approximation to quantification | |
| Cappellari et al. | Resolution scalable image coding with reversible cellular automata | |
| Bernabé et al. | A lossy 3D wavelet transform for high-quality compression of medical video | |
| Cai et al. | Smart wavelet image coding: X-tree approach | |
| Jakisc et al. | Analysis of different influence of compression algorithm on the image filtered Laplacian, Prewitt and Sobel operator | |
| Shu et al. | Shape adaptive texture coding based on wavelet-based contourlet transform | |
| Ramaswamy et al. | Context-based lossless image coding using EZW framework | |
| Liu et al. | Low bit-rate image coding based on pyramidal directional filter banks | |
| Song et al. | Contourlet image coding based on adjusted SPIHT | |
| Shu et al. | Arbitrarily shape object-based video coding technology by wavelet-based contourlet transform | |
| Raghuwanshi et al. | A Review of Image Compression based on Wavelet Transform Function and Structure optimisation Technique | |
| Muzaffar et al. | Linked significant tree wavelet-based image compression | |
| CN103517082B (en) | The conversion method and system of jpeg image and wavelet compressed images | |
| Kaur et al. | Listless Block Tree Coding with Discrete Wavelet Transform for Embedded Image Compression at Low Bit Rate | |
| Ghodhbani et al. | Analysis and implementation of parallel causal bit plane coding in JPEG2000 standard | |
| Shwetha et al. | Analysis of Image Compression Algorithms in WSN: A Review | |
| El-Sharkawey et al. | Comparison between (RLE & Huffman and DWT) Algorithms for Data Compression |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C12 | Rejection of a patent application after its publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20130403 |