CN107170020A - Dictionary learning still image compression method based on minimum quantization error criterion - Google Patents
Dictionary learning still image compression method based on minimum quantization error criterion Download PDFInfo
- Publication number
- CN107170020A CN107170020A CN201710417963.5A CN201710417963A CN107170020A CN 107170020 A CN107170020 A CN 107170020A CN 201710417963 A CN201710417963 A CN 201710417963A CN 107170020 A CN107170020 A CN 107170020A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- dictionary
- mtd
- msubsup
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T9/00—Image coding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
- H03M7/00—Conversion of a code where information is represented by a given sequence or number of digits to a code where the same, similar or subset of information is represented by a different sequence or number of digits
- H03M7/30—Compression; Expansion; Suppression of unnecessary data, e.g. redundancy reduction
- H03M7/3059—Digital compression and data reduction techniques where the original information is represented by a subset or similar information, e.g. lossy compression
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
- H03M7/00—Conversion of a code where information is represented by a given sequence or number of digits to a code where the same, similar or subset of information is represented by a different sequence or number of digits
- H03M7/30—Compression; Expansion; Suppression of unnecessary data, e.g. redundancy reduction
- H03M7/60—General implementation details not specific to a particular type of compression
- H03M7/6041—Compression optimized for errors
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
- H03M7/00—Conversion of a code where information is represented by a given sequence or number of digits to a code where the same, similar or subset of information is represented by a different sequence or number of digits
- H03M7/30—Compression; Expansion; Suppression of unnecessary data, e.g. redundancy reduction
- H03M7/60—General implementation details not specific to a particular type of compression
- H03M7/6064—Selection of Compressor
- H03M7/6082—Selection strategies
- H03M7/6088—Selection strategies according to the data type
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Probability & Statistics with Applications (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The invention discloses a kind of dictionary learning still image compression method based on minimum quantization error criterion, the technical problem big for solving existing still image compression method quantization error.Technical scheme is added the comentropy of sparse coefficient manipulative indexing as regular terms in the object function of sparse coding, the dictionary atomic time is being chosen using orthogonal matching pursuit algorithm, the decentralization of dictionary atom is limited by minimizing comentropy, the Coding cost of sparse coefficient manipulative indexing is reduced;Simultaneously, during dictionary learning, by being ranked up to sparse coefficient, and find the used sequence divisions of the k for causing the total sum of squares of deviations of sparse coefficient minimum, to each it divide as a quantization group, used between different quantization groups and identical quantization step is used in different quantization steps, same quantization group, so that final quantization error is minimum.
Description
Technical field
It is more particularly to a kind of to be based on minimum quantization error criterion the present invention relates to a kind of still image compression method
Dictionary learning still image compression method.
Background technology
Document " Compressibility constrained sparse representation with learnt
Dictionary for low bit-rate image compression, IEEE Transactions on Circuits
And Systems for Video Technology, 2014, Vol24 (10), p1743-1757 " discloses a kind of based on convex pine
Relax and the sparse coding method of compression constraint is used for the lossy compression method of image.This method uses the sparse coding generation based on convex relaxation
Traditional tracking matching algorithm has been replaced, the openness and stability of image representation coefficients is enhanced.Meanwhile, compression constraint is added
Into the solution procedure of sparse coding, sparse coding problem is converted intoNorm optimization problem, this is approached by loop iteration
The optimal solution of problem, so as to obtain sparse coefficient of the image on given super complete dictionary.Finally, by entering to sparse coefficient
Row quantifies and entropy code obtains the compression image code stream of low bit- rate.The dictionary that document methods described is chosen on super complete dictionary is former
Son is dispersed in whole dictionary space so that the Global Information entropy of dictionary atom is higher, it is difficult to carry out efficient coding compression;Separately
Outside, in document methods described, quantization table is learnt using K-means algorithms, and the process is independently of dictionary learning process, therefore
Quantization table can not the super complete dictionary that arrives of adaptive learning, cause quantization error to become big.
The content of the invention
In order to overcome the shortcomings of that existing still image compression method quantization error is big, the present invention provides a kind of based on most
The dictionary learning still image compression method of quantisation errors criterion.This method is by the comentropy of sparse coefficient manipulative indexing
In the object function that sparse coding is added as regular terms, the dictionary atomic time is being chosen using orthogonal matching pursuit algorithm, is being passed through
Minimize comentropy to limit the decentralization of dictionary atom, reduce the Coding cost of sparse coefficient manipulative indexing;Meanwhile, in dictionary
During study, by being ranked up to sparse coefficient, and the used sequences of the k for causing the total sum of squares of deviations of sparse coefficient minimum are found
Divide, will each divide as a quantization group, used between different quantization groups in different quantization steps, same quantization group
Using identical quantization step, so that final quantization error is minimum.
The technical solution adopted for the present invention to solve the technical problems is:A kind of dictionary based on minimum quantization error criterion
Learn still image compression method, be characterized in comprising the following steps:
Step 1: carrying out piecemeal and standardization to training image.All training images are divided into 16 × 16 image block,
For each image block, standardized according to formula (1)
Wherein, vijCoordinate is the gray value of the pixel of (i, j) in expression image block, and m, n represents the length of image block respectively
It is wide.Image block is stretched as vectorConstitute the input signal of dictionary learning.
Step 2: carrying out preliminary clusters to parts of images block first with self-organizing feature map, then pass through K-means algorithms
All image blocks are clustered.In cluster, the distance between any two image block is measured using Euclidean distance
Wherein, si, sjRepresent the different image block vector of any two, d (si, sj) represent its Euclidean distance.
Step 3: training a super complete dictionary and a quantization table, dictionary to each class cluster using dictionary learning algorithm
In each atom be the shared tactic pattern of such cluster image block.Simultaneously by the quantization error of sparse coefficient and its correspondence rope
The comentropy drawn is added as regular terms in the object function of dictionary learning, by iterative formula (3) and formula (5) to amount
Change table and dictionary are learnt simultaneously, reduce final Coding cost.
Wherein, S is original input signal matrix, is often classified as the input signal s after image block is stretchedi, D is dictionary, and A is dilute
Sparse coefficient matrix, αjIt is sparse coefficient matrix A jth row, represents signal sjThe expression coefficient decomposed on dictionary D, kmaxIt is dilute
Dredge degree limitation, piIt is dictionary atom diUse probability, M be dictionary in dictionary atom number.Then, by the word of each class cluster
Allusion quotation and quantization table are spliced into Global Dictionary and global quantization table respectively, are stored in coding side and decoding end.
Step 4: during Image Coding, dividing the image into DC component and AC compounent two parts.DC component is carried out
DPCM is encoded.AC compounent is decomposed on Global Dictionary using sparse coding, corresponding sparse coefficient matrix is obtained, it is right
The sparse coefficient matrix is quantified using global quantization table, then, to the corresponding sparse coefficient square of AC compounent after quantization
Nonzero element and its index carry out Huffman codings in battle array, form final code stream.
Step 5: decoding process is the inverse process of cataloged procedure.It is multiplied by dictionary with sparse coefficient matrix and reconstructs letter
Number matrix S, then to signal matrix each column is plus corresponding DC component and rearranges, so as to recover image.
The beneficial effects of the invention are as follows:This method adds the comentropy of sparse coefficient manipulative indexing as regular terms sparse
In the object function of coding, the dictionary atomic time is being chosen using orthogonal matching pursuit algorithm, is being limited by minimizing comentropy
The decentralization of dictionary atom, reduces the Coding cost of sparse coefficient manipulative indexing;Meanwhile, during dictionary learning, pass through
Sparse coefficient is ranked up, and finds the k for causing the total sum of squares of deviations of sparse coefficient minimum and is used to sequence division, work will be each divided
Used for a quantization group, between different quantization groups in different quantization steps, same quantization group and step is quantified using identical
It is long, so that final quantization error is minimum.
The present invention is elaborated with reference to embodiment.
Embodiment
Dictionary learning still image compression method of the invention based on minimum quantization error criterion is comprised the following steps that:
1. image block and standardization.
All training images are first according to raster scan order, translational movement is 2, is divided into 16 × 16 image blockThen to each image block biProgress is standardized with reference to formula (1)
Finally, image block is stretched as vectorConstitute the input signal of dictionary learning.
2. image block is clustered.
In order to ensure the purity of the dictionary learnt, it is necessary to first cluster image block, make all to be similar diagram in every class
As block.Due to carrying out having overlapping piecemeal to training image, the image number of blocks of acquisition is many, therefore first from all image blocks
In randomly select 10% image block, using Self Organizing Feature Maps Algorithm carry out preliminary clusters, obtain preliminary clusters number of clusters k and
Cluster centreThen by preliminary clusters centerAs initial value, using K-Means algorithms to all image blocks
Clustered.In cluster, the distance between any two image block is measured using Euclidean distance
3. dictionary learning is with quantifying table learning.
In traditional sparse coding algorithm, because signal is decomposed on an excessively complete dictionary, one is rebuild
The linear combination of multigroup different dictionary atom, signal s are there is during signal s1Dictionary atom d can be passed throughiAnd djLinear group
Close s1=α1di+β1djRebuild;Signal s2Both s can be passed through2=α2dp+β2dqRebuild, s can be passed through again2=α2′di+β2′
djRebuild.S in latter reconstruction model1And s2Identical dictionary atom is used, then the selection of dictionary atom is more sparse, compiled
Shorter word length is only needed to during its corresponding index value of code, is conducive to improving compression ratio.In the present invention, dictionary atom is indexed
The comentropy of value is added as regular terms in the object function of sparse coding so that final Coding cost is minimum.It is amended
Object function is with reference to formula (3)
P in formulaiCalculating with reference to formula (4)
Wherein, aiSparse coefficient matrix A the i-th row is represented, δ is a minimum real number, and it is 0 to prevent denominator.
The sparse coefficient matrix A of primary signal matrix is obtained after sparse coding, due to being all floating number in A, it is difficult to
Compression storage, therefore needs to carry out it quantization Q (A), and now needing to update dictionary D again minimizes reconstruction error.Update word
The process of allusion quotation is referred to as dictionary learning, and object function now is with reference to formula (5)
Wherein Q () is quantization function.When quantifying, in order to reduce overall quantization error, using non-uniform quantizing, i.e., pair
Larger sparse coefficient uses big quantization step, and small quantization step is used to less sparse coefficient.In order to realize above-mentioned target,
Nonzero element in sparse coefficient matrix A is expressed as (aij, idxij) form, aijRepresent nonzero element, idxijRepresent that its is right
The index answered.First, to aijCarry out ascending sort and obtain ascending order ordered series of numbers L.Then, ascending order ordered series of numbers L is divided into k used sequence subnumbers
Arrange { L1, L2..., Lk, and make it that the overall sum of squares of deviations after dividing is minimum, that is, meet formula (6)
Wherein,Represent subnumber row LiAverage value, ljRepresent subnumber row LiIn j-th number.Now, k sub- ordered series of numbers pair
Answer to use between k different quantization groups, different quantization groups and identical amount is used in different quantization steps, same quantization group
Change step-length.Eight-digit binary number code e is used during quantization0e1e2e3e4e5e6e7Sparse coefficient is indicated.As k=3, sparse system
Matrix number is divided into eight quantization groups, wherein e0Represent positive and negative, the e of coefficient1e2e3The sparse affiliated quantization group is represented,
e4e5e6e7Represent the dynamic range of the amplitude of the quantization system number.The quantization step of n-th of quantization group is
Wherein,Quantization group L is represented respectivelynThe minimum value and maximum of expression.Coefficient aijQuantized value
It is then
AllThen constitute quantization table.
During whole dictionary learning, formula (3) and formula (5) are optimized by alternating iteration, optimal word is tried to achieve
Allusion quotation D and primary signal matrix sparse coefficient matrix A.The dictionary learnt and quantization table are respectively stored in coding side and decoding
Used when end is for Image Coding, decoding.
4. Image Coding.
During Image Coding, first, image is pressed into raster scan order, be divided into no overlap 16 × 16 image block.Then,
Image block is divided into DC component and AC compounent two parts.DC component is encoded with DPCM.For AC compounent,
Sparse coding is carried out on the dictionary learnt, sparse coefficient matrix is obtained, and it is enterprising in the quantization table learnt to sparse coefficient matrix
Row quantifies.Finally, Huffman codings are carried out to the nonzero element in the sparse coefficient matrix by quantifying and its index, formed
Code stream.
5. image decoding.
During image decoding, first, DC component and sparse coefficient matrix are recovered from code stream.Then, by dictionary with it is dilute
Sparse coefficient matrix multiple reconstructs signal matrix.Finally, to signal matrix each column is plus corresponding DC component and rearranges,
So as to recover image block, original image is recovered by image block splicing.
Claims (1)
1. a kind of dictionary learning still image compression method based on minimum quantization error criterion, it is characterised in that including with
Lower step:
Step 1: carrying out piecemeal and standardization to training image;All training images are divided into 16 × 16 image block, for
Each image block, is standardized according to formula (1)
<mrow>
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<msub>
<mi>v</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</msub>
<mo>=</mo>
<mfrac>
<mrow>
<msub>
<mi>v</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</msub>
<mo>-</mo>
<mi>&mu;</mi>
</mrow>
<msqrt>
<mrow>
<mfrac>
<mn>1</mn>
<mrow>
<mi>m</mi>
<mo>&times;</mo>
<mi>n</mi>
</mrow>
</mfrac>
<msubsup>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>m</mi>
</msubsup>
<msubsup>
<mo>&Sigma;</mo>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</msubsup>
<msup>
<mrow>
<mo>(</mo>
<msub>
<mi>v</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</msub>
<mo>-</mo>
<mi>&mu;</mi>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
</msqrt>
</mfrac>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mi>&mu;</mi>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mrow>
<mi>m</mi>
<mo>&times;</mo>
<mi>n</mi>
</mrow>
</mfrac>
<msubsup>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>m</mi>
</msubsup>
<msubsup>
<mo>&Sigma;</mo>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</msubsup>
<msub>
<mi>v</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</msub>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>,</mo>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, vijCoordinate is the gray value of the pixel of (i, j) in expression image block, and m, n represents the length and width of image block respectively;Will
Image block is stretched as vectorConstitute the input signal of dictionary learning;
Step 2: preliminary clusters are carried out to parts of images block first with self-organizing feature map, then by K-means algorithms to institute
There is image block to be clustered;In cluster, the distance between any two image block is measured using Euclidean distance
<mrow>
<mi>d</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>s</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<msub>
<mi>s</mi>
<mi>j</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mo>|</mo>
<mo>|</mo>
<msub>
<mi>s</mi>
<mi>i</mi>
</msub>
<mo>-</mo>
<msub>
<mi>s</mi>
<mi>j</mi>
</msub>
<mo>|</mo>
<msubsup>
<mo>|</mo>
<mn>2</mn>
<mn>2</mn>
</msubsup>
<mo>;</mo>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>2</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, si, sjRepresent the different image block vector of any two, d (si, sj) represent its Euclidean distance;
Step 3: being trained using dictionary learning algorithm to each class cluster in a super complete dictionary and a quantization table, dictionary
Each atom is the shared tactic pattern of such cluster image block;Simultaneously by the quantization error of sparse coefficient and its manipulative indexing
Comentropy is added as regular terms in the object function of dictionary learning, by iterative formula (3) and formula (5) to quantifying table
Learnt simultaneously with dictionary, reduce final Coding cost;
<mrow>
<mtable>
<mtr>
<mtd>
<mrow>
<mover>
<mi>A</mi>
<mo>^</mo>
</mover>
<mo>=</mo>
<mi>arg</mi>
<mi> </mi>
<msub>
<mi>min</mi>
<mi>A</mi>
</msub>
<mo>{</mo>
<mo>|</mo>
<mo>|</mo>
<mi>S</mi>
<mo>-</mo>
<mi>D</mi>
<mi>A</mi>
<mo>|</mo>
<msubsup>
<mo>|</mo>
<mi>F</mi>
<mn>2</mn>
</msubsup>
<mo>-</mo>
<mi>&lambda;</mi>
<msubsup>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>M</mi>
</msubsup>
<msub>
<mi>p</mi>
<mi>i</mi>
</msub>
<mi>log</mi>
<mi> </mi>
<msub>
<mi>p</mi>
<mi>i</mi>
</msub>
<mo>}</mo>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mtable>
<mtr>
<mtd>
<mrow>
<mi>s</mi>
<mo>.</mo>
<mi>t</mi>
<mo>.</mo>
</mrow>
</mtd>
<mtd>
<mrow>
<mo>|</mo>
<mo>|</mo>
<msub>
<mi>a</mi>
<mrow>
<mo>&CenterDot;</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>|</mo>
<msub>
<mo>|</mo>
<mn>0</mn>
</msub>
<mo>&le;</mo>
<msub>
<mi>k</mi>
<mi>max</mi>
</msub>
<mo>,</mo>
<mn>1</mn>
<mo>&le;</mo>
<mi>j</mi>
<mo>&le;</mo>
<mi>N</mi>
</mrow>
</mtd>
</mtr>
</mtable>
</mtd>
</mtr>
</mtable>
<mo>,</mo>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>3</mn>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mtable>
<mtr>
<mtd>
<mrow>
<mover>
<mi>D</mi>
<mo>^</mo>
</mover>
<mo>=</mo>
<mi>arg</mi>
<mi> </mi>
<msub>
<mi>min</mi>
<mi>D</mi>
</msub>
<mo>|</mo>
<mo>|</mo>
<mi>S</mi>
<mo>-</mo>
<mi>D</mi>
<mo>&CenterDot;</mo>
<mi>Q</mi>
<mrow>
<mo>(</mo>
<mi>A</mi>
<mo>)</mo>
</mrow>
<mo>|</mo>
<msubsup>
<mo>|</mo>
<mi>F</mi>
<mn>2</mn>
</msubsup>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mtable>
<mtr>
<mtd>
<mrow>
<mi>s</mi>
<mo>.</mo>
<mi>t</mi>
<mo>.</mo>
</mrow>
</mtd>
<mtd>
<mrow>
<mo>|</mo>
<mo>|</mo>
<msub>
<mi>a</mi>
<mrow>
<mo>&CenterDot;</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>|</mo>
<msub>
<mo>|</mo>
<mn>0</mn>
</msub>
<mo>&le;</mo>
<msub>
<mi>k</mi>
<mi>max</mi>
</msub>
<mo>,</mo>
<mn>1</mn>
<mo>&le;</mo>
<mi>j</mi>
<mo>&le;</mo>
<mi>N</mi>
</mrow>
</mtd>
</mtr>
</mtable>
</mtd>
</mtr>
</mtable>
<mo>,</mo>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>5</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, S is original input signal matrix, is often classified as the input signal s after image block is stretchedi, D is dictionary, and A is sparse system
Matrix number, a.jIt is sparse coefficient matrix A jth row, represents signal sjThe expression coefficient decomposed on dictionary D, kmaxIt is degree of rarefication
Limitation, piIt is dictionary atom diUse probability, M be dictionary in dictionary atom number;Then, by the dictionary of each class cluster and
Quantization table is spliced into Global Dictionary and global quantization table respectively, is stored in coding side and decoding end;
Step 4: during Image Coding, dividing the image into DC component and AC compounent two parts;DPCM volumes are carried out to DC component
Code;AC compounent is decomposed on Global Dictionary using sparse coding, corresponding sparse coefficient matrix is obtained, it is sparse to this
Coefficient matrix is quantified using global quantization table, then, to non-in the corresponding sparse coefficient matrix of AC compounent after quantization
Neutral element and its index carry out Huffman codings, form final code stream;
Step 5: decoding process is the inverse process of cataloged procedure;It is multiplied by dictionary with sparse coefficient matrix and reconstructs signal square
Battle array S, then to signal matrix each column is plus corresponding DC component and rearranges, so as to recover image.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710417963.5A CN107170020B (en) | 2017-06-06 | 2017-06-06 | Dictionary learning still image compression method based on minimum quantization error criterion |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710417963.5A CN107170020B (en) | 2017-06-06 | 2017-06-06 | Dictionary learning still image compression method based on minimum quantization error criterion |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107170020A true CN107170020A (en) | 2017-09-15 |
CN107170020B CN107170020B (en) | 2019-06-04 |
Family
ID=59825586
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710417963.5A Active CN107170020B (en) | 2017-06-06 | 2017-06-06 | Dictionary learning still image compression method based on minimum quantization error criterion |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107170020B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111274349A (en) * | 2020-01-21 | 2020-06-12 | 北方工业大学 | Public security data hierarchical indexing method and device based on information entropy |
CN113454975A (en) * | 2018-12-13 | 2021-09-28 | 马特瑞勒耶斯公司 | Method, computer program product and system for representing visual information |
CN113922823A (en) * | 2021-10-29 | 2022-01-11 | 电子科技大学 | Social media information propagation graph data compression method based on constraint sparse representation |
WO2024032775A1 (en) * | 2022-08-12 | 2024-02-15 | 华为技术有限公司 | Quantization method and apparatus |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20100014758A1 (en) * | 2008-07-15 | 2010-01-21 | Canon Kabushiki Kaisha | Method for detecting particular object from image and apparatus thereof |
CN102142139A (en) * | 2011-03-25 | 2011-08-03 | 西安电子科技大学 | Compressed learning perception based SAR (Synthetic Aperture Radar) high-resolution image reconstruction method |
CN103489203A (en) * | 2013-01-31 | 2014-01-01 | 清华大学 | Image coding method and system based on dictionary learning |
-
2017
- 2017-06-06 CN CN201710417963.5A patent/CN107170020B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20100014758A1 (en) * | 2008-07-15 | 2010-01-21 | Canon Kabushiki Kaisha | Method for detecting particular object from image and apparatus thereof |
CN102142139A (en) * | 2011-03-25 | 2011-08-03 | 西安电子科技大学 | Compressed learning perception based SAR (Synthetic Aperture Radar) high-resolution image reconstruction method |
CN103489203A (en) * | 2013-01-31 | 2014-01-01 | 清华大学 | Image coding method and system based on dictionary learning |
Non-Patent Citations (6)
Title |
---|
JULIEN MAIRAL等: "Online Dictionary Learning for Sparse Coding", 《PROCEEDINGS OF THE 26 TH INTERNATIONAL CONFERENCE》 * |
MAI XU等: "Compressibility Constrained Sparse Representation With Learnt Dictionary for Low Bit-Rate Image Compression", 《IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY》 * |
SUJIT KUMAR SAHOO等: "Signal Recovery from Random Measurements via Extended Orthogonal Matching Pursuit", 《IEEE TRANSACTIONS ON SIGNAL PROCESSING》 * |
张勋等: "基于字典学习与稀疏表示的灰度图像颜色重建算法", 《计算机辅助设计与图形学学报》 * |
郑兴明等: "基于字典学习正则化的图像去噪", 《计算机工程》 * |
酉霞等: "基于改进K-SVD字典学习的医学图像压缩算法", 《西南科技大学学报》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113454975A (en) * | 2018-12-13 | 2021-09-28 | 马特瑞勒耶斯公司 | Method, computer program product and system for representing visual information |
CN111274349A (en) * | 2020-01-21 | 2020-06-12 | 北方工业大学 | Public security data hierarchical indexing method and device based on information entropy |
CN113922823A (en) * | 2021-10-29 | 2022-01-11 | 电子科技大学 | Social media information propagation graph data compression method based on constraint sparse representation |
CN113922823B (en) * | 2021-10-29 | 2023-04-21 | 电子科技大学 | Social media information propagation graph data compression method based on constraint sparse representation |
WO2024032775A1 (en) * | 2022-08-12 | 2024-02-15 | 华为技术有限公司 | Quantization method and apparatus |
Also Published As
Publication number | Publication date |
---|---|
CN107170020B (en) | 2019-06-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107170020B (en) | Dictionary learning still image compression method based on minimum quantization error criterion | |
CN107516129B (en) | Dimension self-adaptive Tucker decomposition-based deep network compression method | |
CN106157339A (en) | The animated Mesh sequence compaction algorithm extracted based on low-rank vertex trajectories subspace | |
US11436228B2 (en) | Method for encoding based on mixture of vector quantization and nearest neighbor search using thereof | |
CN104867165B (en) | A kind of method for compressing image based on transform domain down-sampling technology | |
CN108984642A (en) | A kind of PRINTED FABRIC image search method based on Hash coding | |
CN110278444B (en) | Sparse representation three-dimensional point cloud compression method adopting geometric guidance | |
Wu et al. | Learning product codebooks using vector-quantized autoencoders for image retrieval | |
CN107992611A (en) | The high dimensional data search method and system of hash method are distributed based on Cauchy | |
CN108846873A (en) | A kind of Medical Image Lossless Compression method based on gray probability | |
Barbalho et al. | Hierarchical SOM applied to image compression | |
CN116939226A (en) | Low-code-rate image compression-oriented generated residual error repairing method and device | |
CN109712205A (en) | A kind of compression of images perception method for reconstructing based on non local self similarity model | |
CN105260736A (en) | Fast image feature representing method based on normalized nonnegative sparse encoder | |
CN114612716A (en) | Target detection method and device based on adaptive decoder | |
CN112702600B (en) | Image coding and decoding neural network layered fixed-point method | |
WO2022057091A1 (en) | Encoding method, decoding method, encoding device, and decoding device for point cloud attribute | |
CN116600119B (en) | Video encoding method, video decoding method, video encoding device, video decoding device, computer equipment and storage medium | |
CN106331719A (en) | K-L transformation error space dividing based image data compression method | |
CN105718858B (en) | A kind of pedestrian recognition method based on positive and negative broad sense maximum pond | |
CN116843830A (en) | Mask image modeling algorithm based on self-supervision learning | |
CN110349228B (en) | Triangular mesh compression method for data-driven least square prediction | |
Kekre et al. | Vector quantized codebook optimization using modified genetic algorithm | |
Zhu et al. | Learning low-rank representations for model compression | |
CN116740414B (en) | Image recognition method, device, electronic equipment and storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |