CN101986630B - Spherical vector quantization coding and decoding method - Google Patents
Spherical vector quantization coding and decoding method Download PDFInfo
- Publication number
- CN101986630B CN101986630B CN2010105112448A CN201010511244A CN101986630B CN 101986630 B CN101986630 B CN 101986630B CN 2010105112448 A CN2010105112448 A CN 2010105112448A CN 201010511244 A CN201010511244 A CN 201010511244A CN 101986630 B CN101986630 B CN 101986630B
- Authority
- CN
- China
- Prior art keywords
- sphere
- companding
- vector
- quantization
- code book
- 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
Images
Abstract
The invention belongs to the technical field of quantization coding, in particular to a spherical vector quantization coding and decoding method. In the method, a spherical quantization codebook is used for participating in the quantization, wherein the spherical quantization codebook is obtained by carrying out spherical truncation in a lattice vector quantization codebook in D-dimension space, and the spherical quantization codebook comprises a peripheral sphere layer which is composed of space lattice points on and near a certain hypersphere or sphere in the D-dimension space, wherein D is more than or equal to 2; and when D is equal to 2, the sphere is degenerated to form a ring. The sphere quantization codebook is used for quantizing vector data which meets Gaussian distribution, thus the quality of the quantization is good, and the efficiency of the quantization is high.
Description
Technical field
The present invention relates to the quantization encoding technical field, more specifically, relate to sphere vector quantization decoding method to gaussian distribution data.
Background technology
In quantization encoding, vector quantization method can obtain the quantizing distortion lower than mark quantization methods under identical number of coded bits.The lattice vector quantization device is to come signal is quantized as vector quantizer with the lattice point in the space lattice.It has advantages such as low memory space, low computational complexity, high quantization precision, has therefore become the technology that extensively adopt in current mobile audio coding decoding field.Have the scholar to prove in theory, lattice vector quantization cooperates the entropy coding based on probability can obtain best vector quantizer, and therefore, 2003 so far, becomes the focus of research based on the lattice vector quantization of probabilistic model.
A kind of common methods that probabilistic model is applied to lattice vector quantization is, uses certain mapping to handle importing data to be quantified, makes it meet certain specific distribution, uses the lattice vector quantization device that meets this distribution to quantize then.Wherein, Gaussian distribution is common a kind of distribution, adopts the ball-type vector quantizer usually.Yet owing to satisfy in the data of Gaussian distribution, little energy vectors has accounted for the overwhelming majority, and ball-type quantizes to adopt uniform lattice, and relative error is bigger when quantizing little energy vectors.Therefore for the data that satisfy Gaussian distribution, traditional lattice vector quantization device performance still remains to be improved.
Summary of the invention
The present invention is directed to the deficiency of existing quantization method, propose a kind of sphere vector quantization technology scheme of suitable gaussian distribution data, purpose is the quantification performance that accounts for most little energy vectors through raising, realizes that the entire amount voltinism can be to improve.
Sphere vector quantization coding method technical scheme provided by the invention adopts sphere to quantize code book and participates in quantizing; Said sphere quantizes code book and from the lattice vector quantization code book of D dimension space, carries out sphere and block and obtain; Comprise by on certain hypersphere in the D dimension space or the sphere and near the peripheral ball layer formed of space lattice; Peripheral ball layer inside comprises hollow region, and wherein D is more than or equal to 2, and sphere is degenerated to ring when D equals 2; Cataloged procedure may further comprise the steps,
Step 1.1, the input vector to be quantified of D dimension is carried out companding, obtain vector and corresponding companding parameter behind the companding in sphere quantizes the code book scope;
Step 1.2, gained companding parameter in the step 1.1 is quantized, obtain the quantization index of companding parameter;
Step 1.3, to vector behind the gained companding in the step 1.1, use sphere to quantize code book it quantized, obtain the quantization index of vector behind the companding;
Step 1.4, with the quantization index of vector behind the step 1.2 gained companding, and the quantization index of the corresponding companding parameter of step 1.3 gained writes code stream.
And said sphere quantizes code book and at the center of hypersphere or sphere astrosphere is set, and is hollow region between peripheral ball layer and the astrosphere.
Corresponding sphere vector quantization decoder method and technology scheme provided by the invention adopts sphere to quantize code book and participates in inverse quantization; Said sphere quantizes code book and from the lattice vector quantization code book of D dimension space, carries out sphere and block and obtain; Comprise by on certain hypersphere in the D dimension space or the sphere and near the peripheral ball layer formed of space lattice; Peripheral ball layer inside comprises hollow region, and wherein D is more than or equal to 2, and sphere is degenerated to ring when D equals 2; Decode procedure may further comprise the steps,
Step 2.1, from code stream, extract the quantization index of vector behind the companding and the quantization index of corresponding companding parameter;
Step 2.2, according to the quantization index of vector behind the step 2.1 gained companding, quantize vector behind the companding after code book draws corresponding quantitative through searching sphere;
Step 2.3, according to the quantization index of step 2.1 gained companding parameter, through searching the companding parameter after companding parameter quantification code book obtains quantizing;
Step 2.4, the companding parameter after quantizing according to step 2.3 gained, vector behind the companding after step 2.2 gained quantized is rebuild the vector to be quantified after quantizing through contrary companding, and the vector to be quantified after the quantification is the D dimension.
And said sphere quantizes code book and at the center of hypersphere or sphere astrosphere is set, and is hollow region between peripheral ball layer and the astrosphere.
Technical scheme of the present invention adopts sphere vector quantization scheme to quantize to the vector data that satisfies Gaussian distribution, has improved quantification quality and efficient, and corresponding inverse quantization scheme, practical simplicity are provided.
Description of drawings
Fig. 1 is that sphere of the present invention quantizes the code book sketch map;
Fig. 2 is the sphere vector quantization apparatus work sketch map of the embodiment of the invention;
Fig. 3 is the sphere vector re-quantization device work sketch map of the embodiment of the invention.
Embodiment
The invention provides sphere vector quantization coding method to gaussian distribution data, and corresponding sphere vector quantization decoder method.During practical implementation, can adopt the computer software means to realize the automation quantified coding by those skilled in the art according to providing technical scheme.Because in encoding and decoding are used, toward contact the encoding and decoding software approach is solidify to form coding and decoding device, the present invention also provides corresponding sphere vector quantization apparatus and sphere vector re-quantization device.
Elaborate below in conjunction with accompanying drawing and embodiment.
Embodiment implements under the inventive method prerequisite, provided detailed execution mode and concrete operating process, but realization of the present invention is not limited to this embodiment.
Usually the code book of lattice vector quantization generally all is ball-type or pyramid; Because the lattice point that is covered with the D dimension space that the code book of lattice vector quantization is to use mathematical method to generate; Quantizing bit number is limited in the practical application, thereby need block code book.Ball-type blocks, and adopts hypersphere exactly, from code book intercepting D dimension ball.Pyramid is blocked, and uses hypercone, and just D dimension awl blocks code book.The present invention has then creatively designed sphere and has quantized code book.Said sphere quantizes code book and from the lattice vector quantization code book of D dimension space, carries out sphere and block and obtain; Comprise by on certain hypersphere in the D dimension space (3 dimensions are called hypersphere when above) or the sphere and near the peripheral ball layer formed of space lattice; Wherein D is more than or equal to 2, and sphere is degenerated to ring when D equals 2.Peripheral ball layer can include only a lattice point on the face, also can comprise on the face and lattice point on every side, allows that near the space lattice hypersphere or the sphere adds peripheral ball layer, and being equivalent to sphere can have certain thickness.In other words, peripheral ball layer can comprise more than one deck lattice point.But can not be solid in the sphere, that is to say that peripheral ball layer inside must comprise hollow region.Consider some situation; If the companding parameter like correspondence adopts gain factor; Will face infinitely-great problem, the present invention further proposes, and can astrosphere be set at the center at hypersphere or sphere; This moment, hollow region was between peripheral ball layer and the astrosphere cross section of 3 dimension space Internal Spherical Surface as shown in Figure 1.Like this, critical value can be provided, avoid to reach on the companding parameter theory infinitely great, cause the situation that improves data transfer rate greatly.Rule of thumb being set by those skilled in the art when the radius of astrosphere can practical implementation, can comprise some little value points, the most simply is directly to be set at 1 point, i.e. the central point of hypersphere or sphere.It is consistent participating in quantizing to quantize code book with the sphere of inverse quantization, and its Spatial Dimension D is the same with the dimension of input vector to be quantified during enforcement.
Encoding scheme:
Step 1.1, carry out companding, obtain vector (2) and corresponding companding parameter (3) behind the companding importing vector to be quantified (1).The main purpose of companding is with the vector to be quantified in sphere quantizes the code book scope not, carries out conversion through certain algorithm, and the vector to be quantified behind the companding is quantized within the code book scope at sphere.The target of general existing companding is all dwindled, and the present invention is when implementing companding, for the vector to be quantified in sphere quantizes the code book scope not is compressed in the scope, the effect of dwindling and amplifying is provided simultaneously.
The method that can adopt when implementing companding has a lot, use gain factor, linear/non-linear conversion, various Function Mapping all can, for example adopt logarithmic function to carry out companding.Embodiment has taked a kind of linear companding method based on the vector energy, ties up vector X to be quantified, wherein X to the D of each input
iBe the i dimension component of X, i=1,2 ..., D.Calculate companding parameter G through the energy that calculates vector X to be quantified, treat quantization vector X and carry out companding, obtain companding after vector X computing formula following:
For the purpose of economizing on resources, in fact the just input vector to be quantified in sphere quantizes the code book scope own can not carry out the companding operation, directly carries out next step.The embodiment processing mode does, whether judges earlier input vector to be quantified before the companding at astrosphere, if otherwise carry out companding; If, vector behind the companding is made as zero vector, corresponding companding parameter is 1, carries out next step.Participate in companding at the input of peripheral ball layer vector to be quantified, program is easier like this, and the companding parameter of calculating generally can be 1.
Step 1.2, gained companding parameter (3) in the step 1.1 is quantized, obtain the quantization index (4) of companding parameter.
Embodiment adopts companding parameter quantification code book that companding parameter G to be quantified is quantized; This code book can be obtained by training; The quantization index KG of the companding parameter
after searching code book and obtaining quantizing, or the encoded K G ' of quantization index KG.
Step 1.3, to vector (2) behind the gained companding in the step 1.1, use sphere to quantize code book it quantized, obtain the quantization index (5) of vector behind the companding.
Near the lattice point that embodiment adopts on certain sphere in the D dimension space or sphere is quantizes code book as sphere; Through search for nearest code word as companding after the quantized result
of vector
obtain corresponding quantitative index K; Perhaps with quantization index K with encoded such as entropy codings after, the encoded K of output quantization index K '.
Step 1.4, with the quantization index (5) of vector behind the companding of step 1.2 gained, and the quantization index (4) of the companding parameter of the correspondence of step 1.2 gained writes code stream.
The pairing quantization index K of vector
behind the companding after embodiment quantizes step 1.3 gained, the perhaps encoded K of quantization index K ' write code stream; With the pairing quantization index KG of companding parameter
after the quantification of step 1.2 gained, perhaps the encoded K G ' of quantization index KG writes code stream.
Decoding scheme:
Step 2.1, from code stream, extract the quantization index (5) of vector behind the companding and the quantization index (4) of corresponding companding parameter.
Embodiment extracts the required parameter of this programme decoding from code stream; Write the quantization index of code stream when promptly encoding; Comprise: the quantization index K of vector behind the companding after the quantification
or the encoded K of quantization index K ', and the quantization index KG of the companding parameter
after quantizing or the encoded K G ' of quantization index KG.
Step 2.2, according to the quantization index (5) of vector behind the step 2.1 gained companding, quantize vector (6) behind the companding after code book draws corresponding quantitative through searching sphere;
Embodiment according to the companding after quantizing after the quantization index K of vector
search sphere and quantize code book; Vector
is obvious after obtaining the companding after the corresponding quantitative, during decoding used sphere quantization code this during with coding the used sphere quantization code of step 1.2 this is consistent.If decoding end step 2.1 obtain be quantization index encoded K ', decoding earlier obtains quantization index K and gets final product.
Step 2.3, according to the quantization index (4) of step 2.1 gained companding parameter, through searching the companding parameter (7) after companding parameter quantification code book obtains quantizing;
Embodiment is according to the quantization index KG of the companding parameter
after quantizing; Through searching companding parameter quantification code book; Companding parameter after obtaining quantizing
is if decoding end obtains in step 2.1 is the encoded K G ' of quantization index KG, and decoding earlier obtains quantization index KG earlier and gets final product.Obviously, this code book is consistent with the code book that quantification companding parameter in the step 1.2 is used.
Companding parameter (7) after step 2.4, the quantification that obtains according to step 2.3 to vector (6) behind the companding after the quantification of step 2.2 gained, is rebuild the vector to be quantified (8) after quantizing through contrary companding.
During practical implementation, the companding contrary in contrary companding process and the cataloged procedure gets final product.Among the embodiment, directly quantize for the vector that falls into astrosphere during coding and establish companding parameter 1, need not special in addition the processing during decoding.
The companding parameter
of embodiment after according to the quantification that obtains in the step 2.3 carried out contrary companding to vector behind the companding after the quantification of step 2.2 gained
, and the vector to be quantified after obtaining quantizing
computational methods are following:
Sphere vector quantization apparatus provided by the present invention realizes that with above-mentioned sphere vector quantization coding method principle is consistent, and sphere vector re-quantization device realizes that with above-mentioned sphere vector quantization decoder method principle is consistent.Reference during for ease of hardware enforcement, the generator structure is explained as follows:
Corresponding sphere vector quantization apparatus comprises following part:
Vector companding module (9) is used for vector to be quantified (1) to input and carries out companding and handle, and obtains vector (2) behind the companding, outputs to sphere quantization modules (10), and the companding parameter (3) of correspondence is outputed to companding parameter quantification module (11);
Companding parameter quantification module (11) is used for the companding parameter (3) of input is quantized, and obtains the quantization index (4) of companding parameter, and outputs to code stream;
Sphere quantization modules (10) is used for adopting sphere to quantize code book to vector (2) behind the companding of input and quantizes, and obtains the quantization index (5) of vector behind the companding, and outputs to code stream.
Corresponding sphere vector re-quantization device comprises following part:
Sphere inverse quantization module (12), the quantization index (5) of vector behind the companding that is used for obtaining according to code stream quantizes vector (6) behind the companding after code book obtains corresponding quantitative through searching sphere, and outputs to contrary companding module (14);
Companding parameter inverse quantization module (13), the quantization index (4) of the companding parameter that is used for obtaining according to code stream, re-quantization obtains the companding parameter (7) after the corresponding quantitative, and outputs to contrary companding module (14);
Contrary companding module (14) is used for according to the companding parameter (7) after quantizing, to the contrary companding of vector (6) warp row behind the companding after quantizing, the vector to be quantified (8) after the quantification that obtains rebuilding.
Implement reference for ease of those skilled in the art, the present invention provides more detailed embodiment following with device form.The D n dimensional vector n of embodiment input is 4 n dimensional vector ns, and companding parameter sphere quantizes code book and comprises 2 code words, the E when calculating the companding parameter
REF=20.
The sphere vector quantization apparatus course of work of embodiment, referring to Fig. 2:
(a) calculate 4 dimension inputs vector X={3.2 to be quantified ,-4.2,0, the energy E of 1.6}, E=28.88.Calculate companding parameter G according to formula, obtain G=1.202.
Calculate vector X={2.662 behind the companding to be quantified ,-3.494,0,1.331} according to formula.
(b) the companding parameter G=1.202 that (a) is obtained; Use companding parameter quantification code book to quantize, the companding parameter after obtaining quantizing
also obtains corresponding quantitative index KG.
(c) vector
uses sphere to quantize code book to quantize behind the companding to be quantified that (a) is obtained, and vector behind the companding after obtaining quantizing
obtains corresponding quantitative index K.
(d) the quantization index K with vector
behind the companding after the quantification of gained in (c) writes code stream, or to K back gained encoded K ' the write code stream of encoding.And the quantization index KG of the companding parameter
after the quantification that (b) obtained, again quantization index KG is write code stream.Perhaps quantization index KG is encoded, G ' writes code stream encoded K.
The sphere vector re-quantization device course of work of embodiment, referring to Fig. 3:
(A) encoded K of the quantization index K of vector or quantization index behind the companding after extraction quantizes from code stream ', and the quantization index KG of the companding parameter G after the corresponding quantitative or the encoded K G ' of quantization index KG.
(B) the quantization index KG of the companding parameter
after the basis quantification of (A) gained; Through searching companding parameter quantification code book; Companding parameter after obtaining quantizing
is obvious, and this code book is consistent with the code book that quantification companding parameter in the coding section is used.If that decoding end obtains is the encoded K G ' of quantization index KG, also need passes through decoding and obtain quantization index KG earlier.
(C) the quantization index K of vector behind the companding after basis (A) gained quantizes; Vector
is obvious after searching sphere and quantizing the companding after code book draws corresponding quantitative, during decoding used sphere quantization code this during with coding the used sphere quantization code of step 1.3 this is consistent.If decoding end obtains be quantization index encoded K ', decoding earlier obtains quantization index K.
(D) according to the companding parameter
after the quantification that obtains in (B) vector behind the companding after the quantification that obtains in (C)
is carried out re-quantization according to formula; When rebuilding vector to be quantified
practical application after quantizing; There is error in decoding gained the possibility of result; With vector X={3.2 to be quantified;-4.2; 0,1.6} is not quite identical.But this error common situation that is this area, and error of the present invention is less than the prior art scheme.
Claims (4)
1. sphere vector quantization coding method; It is characterized in that: adopt sphere to quantize code book and participate in quantizing; Said sphere quantizes code book and from the lattice vector quantization code book of D dimension space, carries out sphere and block and obtain, and comprises that peripheral ball layer inside comprises hollow region by on certain hypersphere in the D dimension space or the sphere and near the peripheral ball layer formed of space lattice; Wherein D is more than or equal to 2, and sphere is degenerated to ring when D equals 2; Cataloged procedure may further comprise the steps,
Step 1.1, the input vector to be quantified of D dimension is carried out companding, obtain vector and corresponding companding parameter behind the companding in sphere quantizes the code book scope;
Step 1.2, gained companding parameter in the step 1.1 is quantized, obtain the quantization index of companding parameter;
Step 1.3, to vector behind the gained companding in the step 1.1, use sphere to quantize code book it quantized, obtain the quantization index of vector behind the companding;
Step 1.4, with the quantization index of vector behind the step 1.3 gained companding, and the quantization index of the corresponding companding parameter of step 1.2 gained writes code stream.
2. sphere vector quantization coding method as claimed in claim 1 is characterized in that: said sphere quantizes code book and at the center of hypersphere or sphere astrosphere is set, and is hollow region between peripheral ball layer and the astrosphere.
3. sphere vector quantization decoder method; It is characterized in that: adopt sphere to quantize code book and participate in inverse quantization; Said sphere quantizes code book and from the lattice vector quantization code book of D dimension space, carries out sphere and block and obtain, and comprises that peripheral ball layer inside comprises hollow region by on certain hypersphere in the D dimension space or the sphere and near the peripheral ball layer formed of space lattice; Wherein D is more than or equal to 2, and sphere is degenerated to ring when D equals 2; Decode procedure may further comprise the steps,
Step 2.1, from code stream, extract the quantization index of vector behind the companding and the quantization index of corresponding companding parameter;
Step 2.2, according to the quantization index of vector behind the step 2.1 gained companding, quantize vector behind the companding after code book draws corresponding quantitative through searching sphere;
Step 2.3, according to the quantization index of step 2.1 gained companding parameter, through searching the companding parameter after companding parameter quantification code book obtains quantizing;
Step 2.4, the companding parameter after quantizing according to step 2.3 gained, vector behind the companding after step 2.2 gained quantized is rebuild the vector to be quantified after quantizing through contrary companding, and the vector to be quantified after the quantification is the D dimension.
4. sphere vector quantization decoder method as claimed in claim 3 is characterized in that: said sphere quantizes code book and at the center of hypersphere or sphere astrosphere is set, and is hollow region between peripheral ball layer and the astrosphere.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2010105112448A CN101986630B (en) | 2010-10-18 | 2010-10-18 | Spherical vector quantization coding and decoding method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2010105112448A CN101986630B (en) | 2010-10-18 | 2010-10-18 | Spherical vector quantization coding and decoding method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN101986630A CN101986630A (en) | 2011-03-16 |
CN101986630B true CN101986630B (en) | 2012-11-21 |
Family
ID=43710934
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2010105112448A Expired - Fee Related CN101986630B (en) | 2010-10-18 | 2010-10-18 | Spherical vector quantization coding and decoding method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN101986630B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
PL3467825T3 (en) | 2014-07-28 | 2020-08-10 | Telefonaktiebolaget Lm Ericsson (Publ) | Pyramid vector quantizer shape search |
CN113382418B (en) * | 2021-06-09 | 2023-03-14 | 上海移远通信技术股份有限公司 | Beam codebook generating method and device, mobile terminal and storage medium |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101266796A (en) * | 2007-03-12 | 2008-09-17 | 华为技术有限公司 | A quantified coding method and device |
CN101266795A (en) * | 2007-03-12 | 2008-09-17 | 华为技术有限公司 | An implementation method and device for grid vector quantification coding |
CN101340582A (en) * | 2008-08-13 | 2009-01-07 | 武汉大学 | Motion vector synthesizing method in resolution decreasing video code conversion |
-
2010
- 2010-10-18 CN CN2010105112448A patent/CN101986630B/en not_active Expired - Fee Related
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101266796A (en) * | 2007-03-12 | 2008-09-17 | 华为技术有限公司 | A quantified coding method and device |
CN101266795A (en) * | 2007-03-12 | 2008-09-17 | 华为技术有限公司 | An implementation method and device for grid vector quantification coding |
CN101340582A (en) * | 2008-08-13 | 2009-01-07 | 武汉大学 | Motion vector synthesizing method in resolution decreasing video code conversion |
Also Published As
Publication number | Publication date |
---|---|
CN101986630A (en) | 2011-03-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Huang et al. | 3d point cloud geometry compression on deep learning | |
CN110378468B (en) | Neural network accelerator based on structured pruning and low bit quantization | |
CN102460513B (en) | Method for encoding/decoding a 3D mesh model that comprises one or more components | |
CN1659785B (en) | Method and system for multi-rate lattice vector quantization of a signal | |
CN111818346B (en) | Image encoding method and apparatus, image decoding method and apparatus | |
US11494948B2 (en) | Point cloud geometric compression method based on depth auto-encoder | |
Alvar et al. | Multi-task learning with compressible features for collaborative intelligence | |
CN103957015B (en) | Nonuniform quantizing coding method used for decoding LDPC code and application of method in decoder | |
WO2021023206A1 (en) | Point cloud attribute prediction, encoding, and decoding method and device based on neighbor weight optimization | |
CN113163203B (en) | Deep learning feature compression and decompression method, system and terminal | |
CN103024392B (en) | Method and device for intra-frame mode prediction based on two-dimensional Hadamard transformation | |
CN103716623B (en) | Video compression encoding-and-decoding method and encoder-decoder on the basis of weighting quantification | |
CN101986630B (en) | Spherical vector quantization coding and decoding method | |
Yu et al. | Low-bit quantization needs good distribution | |
Boo et al. | Fixed-point optimization of transformer neural network | |
Akbari et al. | Learned multi-resolution variable-rate image compression with octave-based residual blocks | |
Zhe et al. | Rate-distortion optimized coding for efficient cnn compression | |
CN107070590B (en) | WSN perception data distributed decoding method based on MapReduce | |
Matsuda et al. | Lossless coding using predictors and arithmetic code optimized for each image | |
CN110349228B (en) | Triangular mesh compression method for data-driven least square prediction | |
CN115983343A (en) | YOLOv4 convolutional neural network lightweight method based on FPGA | |
CN101695000A (en) | Disintegration vector quantification coding and decoding method and device thereof based on Gauss mixing model | |
CN103632673A (en) | Nonlinear quantization method of voice linear prediction model | |
Duan et al. | Differential weight quantization for multi-model compression | |
Ding et al. | Improved frequency table adjusting algorithms for context-based adaptive lossless image coding |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20121121 Termination date: 20161018 |
|
CF01 | Termination of patent right due to non-payment of annual fee |