CN108053455A - A kind of vector quantization method for compressing high spectrum image based on linear prediction decorrelation - Google Patents
A kind of vector quantization method for compressing high spectrum image based on linear prediction decorrelation Download PDFInfo
- Publication number
- CN108053455A CN108053455A CN201711421608.1A CN201711421608A CN108053455A CN 108053455 A CN108053455 A CN 108053455A CN 201711421608 A CN201711421608 A CN 201711421608A CN 108053455 A CN108053455 A CN 108053455A
- Authority
- CN
- China
- Prior art keywords
- mrow
- spectrum
- bands
- msub
- munderover
- 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
- G06T9/004—Predictors, e.g. intraframe, interframe 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Probability & Statistics with Applications (AREA)
- Multimedia (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
- Compression, Expansion, Code Conversion, And Decoders (AREA)
Abstract
The present invention discloses a kind of vector quantization method for compressing high spectrum image based on linear prediction decorrelation, including：Step 1：High spectrum image to be compressed is gathered, all bands of a spectrum of high spectrum image to be compressed are divided into multiple classes using the method for cluster, using the cluster centre of every one kind as the reference spectrum band of generation；Step 2：Bands of a spectrum all in high spectrum image to be compressed are predicted using reference spectrum band, regards reference spectrum band as one group of substrate, bands of a spectrum all in high spectrum image to be compressed is all projected in this group of substrate, pass through these bands of a spectrum of the coefficient prediction of projection；Prediction residual is exactly the result after de-redundancy；Step 3：VQ codings are carried out to prediction residual, complete the compression of high spectrum image to be compressed；Final image is compressed to two parts, the reference spectrum band and projection coefficient of bands of a spectrum prediction algorithm and code book and index value in VQ algorithms.The present invention proposes a kind of method of effective bands of a spectrum de-redundancy, and passes through the validity of experimental verification this method.
Description
Technical field
The invention belongs to Hyperspectral imagery processing field, more particularly to a kind of vector quantization based on linear prediction decorrelation
Method for compressing high spectrum image.
Background technology
High spectrum image is usually all the result being imaged to Same Scene in a bands of a spectrum up to a hundred.High spectrum image is schemed in reflection
Have apparent advantage on the material composition of target as in, nowadays since high-spectral data is in national defence, agricultural, geology etc. by
More and more attention are arrived.However bulk redundancy present in high spectrum image constrains transmission and the place of high spectrum image
Reason, therefore the compression of high spectrum image also becomes a popular research topic.Since there is a large amount of in high spectrum image
Redundancy, these redundancies are divided into redundancy between spatial redundancy and spectrum.Three classes can be summarized as the compression method of high spectrum image：In advance
Survey coding method, the method based on conversion and the method based on vector quantization.Vector quantization is that a kind of theoretically optimal block is compiled
Code algorithm, for the compression method of high spectrum image, VQ is applied primarily to removal bands of a spectrum redundancy.At present for high spectrum image
Bands of a spectrum redundancy removing method mostly use prediction point-by point scheme, this kind of method operand is larger, is directly predicted using bands of a spectrum
Method effect it is unsatisfactory.
The content of the invention
It is an object of the invention to provide a kind of vector quantization Compression of hyperspectral images sides based on linear prediction decorrelation
Method, to solve in the prior art due to, there is redundancy between substantial amounts of spatial redundancy and spectrum, seriously being limited in high spectrum image
The problem of transmission and processing of high spectrum image.The present invention is summed up by the research to high spectrum image in high spectrum image
Redundancy will be significantly stronger than spatial redundancy between spectrum.Therefore the present invention is primarily focused between elimination spectrum in redundancy.It considers simultaneously
Strong correlation between spectrum, this method carry out bands of a spectrum whole prediction to achieve the effect that redundancy preferably eliminating spectrum.
In order to achieve the above object, the present invention is achieved by the following scheme：
A kind of vector quantization method for compressing high spectrum image based on linear prediction decorrelation, comprises the following steps：
Step 1：High spectrum image to be compressed is gathered, all bands of a spectrum of high spectrum image to be compressed are used into the side of cluster
Method is divided into multiple classes, using the cluster centre of every one kind as the reference spectrum band of generation；
Step 2：Bands of a spectrum all in high spectrum image to be compressed are predicted using reference spectrum band, by reference spectrum band
Regard one group of substrate as, bands of a spectrum all in high spectrum image to be compressed are all projected in this group of substrate, pass through the coefficient of projection
Predict these bands of a spectrum；Prediction residual is exactly the result after de-redundancy；
Step 3：VQ codings are carried out to prediction residual, complete the compression of high spectrum image to be compressed；Final image is compressed
For two parts, the reference spectrum band and projection coefficient of bands of a spectrum prediction algorithm and code book and index value in VQ algorithms.
Further, step 1 specifically includes：
Reference spectrum band generating process：
1.1) the cross correlation matrix number between all bands of a spectrum is calculated；
1.2) K-means clusters are carried out using every a line in cross correlation matrix number as feature vector；
1.3) bands of a spectrum are averaging to mutually similar bands of a spectrum to be used as with reference to bands of a spectrum；
The input for being used as clustering algorithm by the cross correlation number vector calculated between all bands of a spectrum in clustering algorithm is special
It levies, the cross correlation number calculating method between arbitrary bands of a spectrum is as follows：
I in formula (1), j represent the spatial position of arbitrary pixel respectively, and M and N represent the length of high spectrum image to be compressed
And width, p and q then represent two different spectral bands；rq(p)=rp(q), rp(q) cross-correlation coefficient of expression bands of a spectrum p and q, I (i, j,
Q) any point in high spectrum image is represented,Represent the average of bands of a spectrum q；
It is by vec by calculating the cross-correlation coefficient between all bands of a spectrum to obtain cross correlation matrix number R, Rq(q=1,
2 ..., P) composition；
vecq={ rq(1),rq(2),...,rq(p),...,rq(P) }, p=1,2 ..., P (2)
Obtain the cross-correlation coefficient vector v ec of arbitrary bands of a spectrumqAfterwards, using K-means algorithms to all cross correlations
Number vector is clustered；
The process of clustering algorithm is as follows：
Input：P cross-correlation coefficient vector v ec of all bands of a spectrumqAnd cluster number K, q=1,2 ..., P；
2.1) K cluster centre is generated at random
2.2) all cross correlation number vectors are calculated to the distance of K cluster centre, each cross correlation number vector is drawn
It is divided into the class belonging to a closest cluster centre；
2.3) average to all cross correlation number vectors of every one kind the cluster centre new as such；
2.4) whenWhen, algorithm stops, outputFor final cluster centre, each cross-correlation at this time
Class belonging to coefficient vector is final classification results, otherwise repeatedly step 2.2)；
Output：The classification results and cluster centre of all input cross correlation number vectors
The cluster result obtained eventually by clustering algorithm is the final classification result of all bands of a spectrum；
Ref is obtained by formula (4)k, k=1,2 ..., K is the final reference bands of a spectrum obtained.
Further, step 2 specifically includes：
When carrying out bands of a spectrum prediction：
Y represents bands of a spectrum to be predicted, ref in formulakExpression reference spectrum band, and wkRepresent throwings of the bands of a spectrum y on the reference spectrum band
Shadow.nkFor noise, by solving wkObtain predictive coefficient.
Further, the optimization problem constrained by solution formula (6) and (7) obtains final predictive coefficient；
Wherein X is represented by reference spectrum band refkThe matrix of composition, X={ ref1,ref2,...,refK}；
W is predictive coefficient, w=(w1,w2,...,wK), wlsIt is the predictive coefficient obtained by least square method；IpIt is one
A bands of a spectrum to be predicted,It is the prediction result to the bands of a spectrum；By solving this optimization problem, last result is expressed as formula
(8) and formula (9)；
wls=(XTX)-1XTIp (8)
It is to integral function of going down；The result finally predicted is exactly relevant as a result, this is linear pre- bands of a spectrum removal spectrum
Survey process by integrally being predicted between bands of a spectrum, using reference spectrum band it is linear as substrate represent all bands of a spectrum so as to reach removal spectrum between phase
The effect of closing property.
Further, VQ cataloged procedures are specific as follows：
3.1) piecemeal processing is carried out to image；
3.2) code book is generated using LBG algorithm；
3.3) vector v formed to each piece searches the index idx corresponding to immediate piece in code book,
Wherein cdFor the arbitrary code word in code book, code book includes s altogetherbA code word；
Final code book and the final result that index value idx is coding；
The detailed process of LBG algorithm is as follows：
Input：All input picture block v and code book include the number s of code wordb；
4.1) s is generated at randombA initial code wordS=1,2 ..., sb；
4.2) all input picture blocks are calculated to sbEach input picture block is subdivided into closest by the distance of a code word
Code word belonging to class in；
4.3) average to all input picture blocks of every one kind the code word new as such；
Wherein a (s) is representedThe number for the input picture block that affiliated class is included；
4.4) whenWhen, algorithm stops, outputFor final code word, otherwise repeatedly 4.2)；
Output：All code words
Further, it is further comprising the steps of：
Step 4：It is divided into two parts when being decoded to compressed images：It is decoded, is predicted using VQ algorithms
Residual error predicts all bands of a spectrum using reference spectrum band, the final high spectrum image for obtaining reconstruct.
Further, the number K for carrying out K-means cluster generation reference spectrum bands takes 4.
Further, the code word that VQ codings are carried out in step 3 is 4 × 4 block.
Compared with the prior art, the present invention has the following advantages：It can be eliminated by the scheme that bands of a spectrum are integrally predicted between composing
Redundancy has and calculates simpler convenient performance；Related coefficient is used as feature vector when carrying out bands of a spectrum and integrally predicting,
Reduce the calculation amount of cluster；It is more simpler than the mode of conversion effective by way of clustering and obtaining reference spectrum band；Finally with
Relatively low code check can preferably compress high spectrum image.
Description of the drawings
Fig. 1 is flow chart of the present invention.
Fig. 2 is that bands of a spectrum and the correlation results schematic diagram in space are counted on Indian_pines；Wherein Fig. 2 (a) is flat
Equal Spectral correlation；Fig. 2 (b) is the average correlation arranged in the 50th bands of a spectrum.
Fig. 3 is the matrix of all cross-correlation coefficient vector compositions.
Fig. 4 is bands of a spectrum IpProjection result on reference spectrum band.
Fig. 5 is the composition of this method compressed encoding stream.
Fig. 6 is the comparison of the image and original image that are reconstructed after compressing；Wherein Fig. 6 (a) is the 50th bands of a spectrum of artwork；Fig. 6 (b) is
The 50th bands of a spectrum of reconstructed image after compression；
Fig. 7 is the rate distortion curve that compression method of the present invention uses different size of code book size sb.
Specific embodiment
The present invention is described in further detail below in conjunction with the accompanying drawings：
It please refers to Fig.1 to shown in Fig. 7, a kind of vector quantization high spectrum image pressure based on linear prediction decorrelation of the present invention
Contracting method, comprises the following steps：
Step 1：High spectrum image to be compressed is gathered, it, will by having the fact that very strong correlation, such as Fig. 2 between bands of a spectrum
All bands of a spectrum are divided into multiple classes using the method for cluster in high spectrum image to be compressed, so as to generate reference spectrum band.
Reference spectrum band generating process is as follows：
(1) the cross correlation matrix number between all bands of a spectrum is calculated；
(2) K-means clusters are carried out using every a line in cross correlation matrix number as feature vector；
(3) bands of a spectrum are averaging to mutually similar bands of a spectrum to be used as with reference to bands of a spectrum.
The input for being used as clustering algorithm by the cross correlation number vector calculated between all bands of a spectrum in clustering algorithm is special
It levies, the cross correlation number calculating method between arbitrary bands of a spectrum is as follows：
I in formula (1), j represent the spatial position of arbitrary pixel respectively, and M and N represent the length of high spectrum image to be compressed
And width, p and q then represent two different spectral bands.rq(p)=rp(q), rp(q) cross-correlation coefficient of expression bands of a spectrum p and q, I (i, j,
Q) any point in high spectrum image is represented,Represent the average of bands of a spectrum q.
It is by vec by calculating the cross-correlation coefficient between all bands of a spectrum to obtain cross correlation matrix number R, Rq(q=1,
2 ..., P) composition.Such as Fig. 3.
vecq={ rq(1),rq(2),...,rq(p),...,rq(P) }, p=1,2 ..., P (2)
Obtain the cross-correlation coefficient vector v ec of arbitrary bands of a spectrumqAfterwards, using K-means algorithms to all cross correlations
Number vector is clustered.
The process of clustering algorithm is as follows：
Input：P cross-correlation coefficient vector v ec of all bands of a spectrumq(q=1,2 ..., P) and cluster number K.
1) K cluster centre is generated at random
2) all cross correlation number vectors are calculated to the distance of K cluster centre, each cross correlation number vector is divided
Enter in the class belonging to a closest cluster centre.
3) average to all cross correlation number vectors of every one kind the cluster centre new as such.
ε represent one close to 0 minimum, for the iteration convergence condition of limit algorithm.
4) whenWhen, algorithm stops, outputFor final cluster centre, each cross-correlation coefficient at this time
Class belonging to vector is final classification results, otherwise repeatedly step 2).
Output：The classification results and cluster centre of all input cross correlation number vectors
The cluster result obtained eventually by clustering algorithm is the final classification result of all bands of a spectrum.
Ref is obtained by formula (4)k, k=1,2 ..., K is the final reference bands of a spectrum obtained.
Step 2：It is predicted using reference spectrum band, regards reference spectrum band as one group of substrate, will be owned in high spectrum image
Bands of a spectrum all project in this group of substrate, these bands of a spectrum, such as Fig. 4 are predicted by the coefficient of projection.Prediction residual is exactly de-redundant
Result after remaining.When carrying out bands of a spectrum prediction：
Y represents bands of a spectrum to be predicted, ref in formula (5)kExpression reference spectrum band, and wkRepresent bands of a spectrum y on the reference spectrum band
Projection.nkFor noise；
The optimization problem constrained by solution formula (6) and (7) can obtain final predictive coefficient.
Wherein X is represented by reference spectrum band refkThe matrix of composition, X={ ref1,ref2,...,refK}。
W is predictive coefficient, w=(w1,w2,...,wK), wlsIt is the predictive coefficient obtained by least square method.IpIt is one
A bands of a spectrum to be predicted,It is the prediction result to the bands of a spectrum；By solving this optimization problem, last result is expressed as formula
(8) and formula (9).
wls=(XTX)-1XTIp (8)
It is to integral function of going down.The result finally predicted is exactly relevant as a result, this is linear pre- bands of a spectrum removal spectrum
Survey process by integrally being predicted between bands of a spectrum, using reference spectrum band it is linear as substrate represent all bands of a spectrum so as to reach removal spectrum between phase
The effect of closing property.
Step 3：VQ codings are carried out to the prediction residual for having eliminated a large amount of correlations, reach the mesh for promoting coding efficiency
's.Final image is compressed to two parts, the code book in the reference spectrum band and projection coefficient of bands of a spectrum prediction algorithm and VQ algorithms
And index value.Since prediction residual eliminates a large amount of correlations, coding efficiency can be promoted.The code of VQ codings is carried out in step 3
Word is 4 × 4 block, and code book size then makes choice according to the size of bands of a spectrum in high spectrum image to be compressed.The volume of VQ algorithms
Code process is specific as follows：
(1) piecemeal processing is carried out to image, may occur in which between block and block Chong Die.
(2) code book is generated using LBG algorithm.
(3) vector v formed to each piece searches the index idx corresponding to immediate piece in code book,
Wherein cdFor the arbitrary code word in code book, code book includes s altogetherbA code word.
Final code book and the final result that index value idx is coding.
The detailed process of LBG algorithm is as follows：
Input：All input picture block v and code book include the number K of code word.
1) s is generated at randombA initial code word
2) all input picture blocks are calculated to sbEach input picture block is subdivided into closest by the distance of a code word
In class belonging to code word.
3) average to all input picture blocks of every one kind the code word new as such.
Wherein a (s) represents the number of such input picture block included.
4) whenWhen, algorithm stops, outputFor final code word, otherwise repeatedly 2).
Output：All code words
The encoding stream such as Fig. 5 obtained by the step.
Step 4：Compressed images are being decoded and are being reconstructed.The step can be divided into the following manner progress：
(1) it is decoded using VQ algorithms, obtains prediction residual, decoded process is the inverse process of coding, by code
The corresponding code word of index value is searched in book to complete the restructuring procedure of prediction residual.
(2) all bands of a spectrum are predicted using reference spectrum band, the generating process for predicting bands of a spectrum is to utilize reference spectrum band and every
The predicted vector (predictive coefficient) of a bands of a spectrum is come what is calculated.Each dimension of predicted vector is original bands of a spectrum to be predicted right
Answer the projection value in substrate (reference spectrum band).Therefore prediction bands of a spectrum are exactly the process that these projections combine.Formula (12)
Illustrate that this process finally obtains the high spectrum image of reconstruct.
The code check of ultimate method calculates as follows
K is the number of reference spectrum band in formula, this parameter has just been determined when K-means is clustered, sbIt is code book
Size, swIt is the dimension of code word.The encoding stream of final output is made of three parts, is reference spectrum band respectively, and fallout predictor is
Number and code book and index in VQ.
It can be seen that the method for compressing high spectrum image can be preferably completed to high spectrum image by Fig. 6 and Fig. 7
Compression.Fig. 6 (a) represents a bands of a spectrum in high spectrum image, and (b) represents the image recovered after being compressed with method in the invention, I
The method of the present invention can compress image well in the case where not produced bigger effect to picture quality.Fig. 7 represents this hair
Bright method is selecting the contrast effect of different size code book, it can be seen that the distortion performance of the method for the present invention is in 256 code words
Preferable performance has been had been able on code book.
Claims (8)
1. a kind of vector quantization method for compressing high spectrum image based on linear prediction decorrelation, which is characterized in that including following
Step：
Step 1：High spectrum image to be compressed is gathered, all bands of a spectrum of high spectrum image to be compressed are used into the method point of cluster
For multiple classes, using the cluster centre of every one kind as the reference spectrum band of generation；
Step 2：Bands of a spectrum all in high spectrum image to be compressed are predicted using reference spectrum band, reference spectrum band is regarded as
One group of substrate, bands of a spectrum all in high spectrum image to be compressed are all projected in this group of substrate, pass through the coefficient prediction of projection
These bands of a spectrum；Prediction residual is exactly the result after de-redundancy；
Step 3：VQ codings are carried out to prediction residual, complete the compression of high spectrum image to be compressed；Final image is compressed to two
Part, the reference spectrum band and projection coefficient of bands of a spectrum prediction algorithm and code book and index value in VQ algorithms.
2. a kind of vector quantization method for compressing high spectrum image based on linear prediction decorrelation according to claim 1,
It is characterized in that, step 1 specifically includes：
Reference spectrum band generating process：
1.1) the cross correlation matrix number between all bands of a spectrum is calculated；
1.2) K-means clusters are carried out using every a line in cross correlation matrix number as feature vector；
1.3) bands of a spectrum are averaging to mutually similar bands of a spectrum to be used as with reference to bands of a spectrum；
It is used as the input feature vector of clustering algorithm by the cross correlation number vector calculated between all bands of a spectrum in clustering algorithm, appoints
Cross correlation number calculating method between meaning bands of a spectrum is as follows：
<mrow>
<msub>
<mi>r</mi>
<mi>q</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>p</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mrow>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>N</mi>
</munderover>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>M</mi>
</munderover>
<mrow>
<mo>(</mo>
<mi>I</mi>
<mo>(</mo>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
<mo>,</mo>
<mi>p</mi>
</mrow>
<mo>)</mo>
<mo>-</mo>
<mover>
<mrow>
<mi>I</mi>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
<mo>,</mo>
<mi>p</mi>
<mo>)</mo>
</mrow>
</mrow>
<mo>&OverBar;</mo>
</mover>
<mo>)</mo>
</mrow>
<mrow>
<mo>(</mo>
<mi>I</mi>
<mo>(</mo>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
<mo>,</mo>
<mi>p</mi>
</mrow>
<mo>)</mo>
<mo>-</mo>
<mover>
<mrow>
<mi>I</mi>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
<mo>,</mo>
<mi>q</mi>
<mo>)</mo>
</mrow>
</mrow>
<mo>&OverBar;</mo>
</mover>
<mo>)</mo>
</mrow>
</mrow>
<msqrt>
<mrow>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>N</mi>
</munderover>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>M</mi>
</munderover>
<msup>
<mrow>
<mo>(</mo>
<mi>I</mi>
<mo>(</mo>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
<mo>,</mo>
<mi>p</mi>
</mrow>
<mo>)</mo>
<mo>-</mo>
<mover>
<mrow>
<mi>I</mi>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
<mo>,</mo>
<mi>p</mi>
<mo>)</mo>
</mrow>
</mrow>
<mo>&OverBar;</mo>
</mover>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>N</mi>
</munderover>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>M</mi>
</munderover>
<msup>
<mrow>
<mo>(</mo>
<mi>I</mi>
<mo>(</mo>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
<mo>,</mo>
<mi>p</mi>
</mrow>
<mo>)</mo>
<mo>-</mo>
<mover>
<mrow>
<mi>I</mi>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
<mo>,</mo>
<mi>q</mi>
<mo>)</mo>
</mrow>
</mrow>
<mo>&OverBar;</mo>
</mover>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
</msqrt>
</mfrac>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
I in formula (1), j represent the spatial position of arbitrary pixel respectively, M and N represent high spectrum image to be compressed length and
Width, p and q then represent two different spectral bands；rq(p)=rp(q), rp(q) cross-correlation coefficient of bands of a spectrum p and q, I (i, j, q) are represented
Represent any point in high spectrum image,Represent the average of bands of a spectrum q；
It is by vec by calculating the cross-correlation coefficient between all bands of a spectrum to obtain cross correlation matrix number R, Rq(q=1,2 ..., P)
Composition；
vecq={ rq(1),rq(2),...,rq(p),...,rq(P) }, p=1,2 ..., P (2)
P represents bands of a spectrum number；
Obtain the cross-correlation coefficient vector v ec of arbitrary bands of a spectrumqAfterwards, using K-means algorithms to all cross correlation number vectors
It is clustered；
The process of clustering algorithm is as follows：
Input：P cross-correlation coefficient vector v ec of all bands of a spectrumqAnd cluster number K, q=1,2 ..., P；
2.1) K cluster centre is generated at random
2.2) all cross correlation number vectors are calculated to the distance of K cluster centre, each cross correlation number vector is subdivided into
In class belonging to a closest cluster centre；
2.3) average to all cross correlation number vectors of every one kind the cluster centre new as such；
<mrow>
<msubsup>
<mi>C</mi>
<mi>k</mi>
<mi>t</mi>
</msubsup>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>n</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mrow>
<mi>c</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
</mrow>
</munderover>
<msub>
<mi>vec</mi>
<mi>n</mi>
</msub>
<mo>/</mo>
<mi>c</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>3</mn>
<mo>)</mo>
</mrow>
</mrow>
2.4) whenWhen, algorithm stops, outputFor final cluster centre, at this time each cross-correlation coefficient to
Class belonging to amount is final classification results, otherwise repeatedly step 2.2)；
Output：The classification results and cluster centre of all input cross correlation number vectors
The cluster result obtained eventually by clustering algorithm is the final classification result of all bands of a spectrum；
<mrow>
<msub>
<mi>ref</mi>
<mi>k</mi>
</msub>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>n</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mrow>
<mi>c</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
</mrow>
</munderover>
<msup>
<mi>I</mi>
<mi>k</mi>
</msup>
<mrow>
<mo>(</mo>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
<mo>,</mo>
<mi>n</mi>
<mo>)</mo>
</mrow>
<mo>/</mo>
<mi>c</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>4</mn>
<mo>)</mo>
</mrow>
</mrow>
Ref is obtained by formula (4)k, k=1,2 ..., K is the final reference bands of a spectrum obtained.
3. a kind of vector quantization method for compressing high spectrum image based on linear prediction decorrelation according to claim 2,
It is characterized in that, step 2 specifically includes：
When carrying out bands of a spectrum prediction：
<mrow>
<mi>y</mi>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>k</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>K</mi>
</munderover>
<mrow>
<mo>(</mo>
<msub>
<mi>ref</mi>
<mi>k</mi>
</msub>
<mo>&times;</mo>
<msub>
<mi>w</mi>
<mi>k</mi>
</msub>
<mo>+</mo>
<msub>
<mi>n</mi>
<mi>k</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>5</mn>
<mo>)</mo>
</mrow>
</mrow>
Y represents bands of a spectrum to be predicted, ref in formulakExpression reference spectrum band, and wkRepresent projections of the bands of a spectrum y on the reference spectrum band；
nkFor noise, by solving wkObtain predictive coefficient.
4. a kind of vector quantization method for compressing high spectrum image based on linear prediction decorrelation according to claim 3,
It is characterized in that, final predictive coefficient is obtained by the optimization problem that solution formula (6) and (7) constrain；
<mrow>
<msub>
<mi>w</mi>
<mrow>
<mi>l</mi>
<mi>s</mi>
</mrow>
</msub>
<mo>=</mo>
<mi>arg</mi>
<munder>
<mrow>
<mi>m</mi>
<mi>i</mi>
<mi>n</mi>
</mrow>
<mi>w</mi>
</munder>
<mo>|</mo>
<msub>
<mi>I</mi>
<mi>p</mi>
</msub>
<mo>-</mo>
<mover>
<msub>
<mi>I</mi>
<mi>p</mi>
</msub>
<mo>^</mo>
</mover>
<mo>|</mo>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>6</mn>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mtable>
<mtr>
<mtd>
<mrow>
<mi>s</mi>
<mo>.</mo>
<mi>t</mi>
<mo>.</mo>
</mrow>
</mtd>
<mtd>
<mrow>
<mover>
<msub>
<mi>I</mi>
<mi>p</mi>
</msub>
<mo>^</mo>
</mover>
<mo>=</mo>
<msup>
<mi>w</mi>
<mi>T</mi>
</msup>
<mi>X</mi>
</mrow>
</mtd>
</mtr>
</mtable>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>7</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein X is represented by reference spectrum band refkThe matrix of composition, X={ ref1,ref2,...,refK}；
W is predictive coefficient, w=(w1,w2,...,wK), wlsIt is the predictive coefficient obtained by least square method；IpIt is one to treat
Predict bands of a spectrum,It is the prediction result to the bands of a spectrum；By solving this optimization problem, last result is expressed as formula (8)
With formula (9)；
wls=(XTX)-1XTIp (8)
It is to integral function of going down；The result finally predicted is exactly relevant as a result, the linear prediction bands of a spectrum removal spectrum
Journey by integrally being predicted between bands of a spectrum, using reference spectrum band it is linear as substrate represent all bands of a spectrum so as to reaching removal Spectral correlation
Effect.
5. a kind of vector quantization method for compressing high spectrum image based on linear prediction decorrelation according to claim 2,
It is characterized in that, VQ cataloged procedures are specific as follows：
3.1) piecemeal processing is carried out to image；
3.2) code book is generated using LBG algorithm；
3.3) vector v formed to each piece searches the index idx corresponding to immediate piece in code book,
<mrow>
<mi>i</mi>
<mi>d</mi>
<mi>x</mi>
<mo>=</mo>
<mi>arg</mi>
<munder>
<mrow>
<mi>m</mi>
<mi>i</mi>
<mi>n</mi>
</mrow>
<mi>d</mi>
</munder>
<mo>{</mo>
<mo>|</mo>
<mo>|</mo>
<mi>v</mi>
<mo>-</mo>
<msub>
<mi>c</mi>
<mi>d</mi>
</msub>
<mo>|</mo>
<msub>
<mo>|</mo>
<mn>2</mn>
</msub>
<mo>|</mo>
<mi>d</mi>
<mo>=</mo>
<mn>1</mn>
<mo>,</mo>
<mn>2</mn>
<mo>,</mo>
<mo>...</mo>
<mo>,</mo>
<msub>
<mi>s</mi>
<mi>b</mi>
</msub>
<mo>}</mo>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>10</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein cdFor the arbitrary code word in code book, code book includes s altogetherbA code word；
Final code book and the final result that index value idx is coding；
The detailed process of LBG algorithm is as follows：
Input：All input picture block v and code book include the number s of code wordb；
4.1) s is generated at randombA initial code word
4.2) all input picture blocks are calculated to sbEach input picture block is subdivided into closest code by the distance of a code word
In class belonging to word；
4.3) average to all input picture blocks of every one kind the code word new as such；
<mrow>
<msubsup>
<mi>c</mi>
<mi>s</mi>
<mi>t</mi>
</msubsup>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>n</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mrow>
<mi>a</mi>
<mrow>
<mo>(</mo>
<mi>s</mi>
<mo>)</mo>
</mrow>
</mrow>
</munderover>
<msub>
<mi>v</mi>
<mi>n</mi>
</msub>
<mo>/</mo>
<mi>a</mi>
<mrow>
<mo>(</mo>
<mi>s</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>11</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein a (s) represents the number of such input picture block included；
4.4) whenWhen, algorithm stops, outputFor final code word, otherwise repeatedly 4.2)；
Output：All code words
6. a kind of vector quantization method for compressing high spectrum image based on linear prediction decorrelation according to claim 1,
It is it is characterized in that, further comprising the steps of：
Step 4：It is divided into two parts when being decoded to compressed images：It is decoded using VQ algorithms, it is residual to obtain prediction
Difference predicts all bands of a spectrum using reference spectrum band, the final high spectrum image for obtaining reconstruct.
7. a kind of vector quantization method for compressing high spectrum image based on linear prediction decorrelation according to claim 2,
It is characterized in that, the number K for carrying out K-means cluster generation reference spectrum bands takes 4.
8. a kind of vector quantization method for compressing high spectrum image based on linear prediction decorrelation according to claim 5,
It is characterized in that, the block that the code word that VQ codings are carried out in step 3 is 4 × 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711421608.1A CN108053455B (en) | 2017-12-25 | 2017-12-25 | Vector quantization hyperspectral image compression method based on linear prediction decorrelation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711421608.1A CN108053455B (en) | 2017-12-25 | 2017-12-25 | Vector quantization hyperspectral image compression method based on linear prediction decorrelation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108053455A true CN108053455A (en) | 2018-05-18 |
| CN108053455B CN108053455B (en) | 2020-06-19 |
Family
ID=62131107
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711421608.1A Active CN108053455B (en) | 2017-12-25 | 2017-12-25 | Vector quantization hyperspectral image compression method based on linear prediction decorrelation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108053455B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114119787A (en) * | 2021-11-30 | 2022-03-01 | 哈尔滨工业大学 | Hyper-spectral image prediction compression method based on orthogonal matching pursuit |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103020912A (en) * | 2012-12-18 | 2013-04-03 | 武汉大学 | Remote sensing image restoration method combining wave-band clustering with sparse representation |
| CN103985096A (en) * | 2014-05-18 | 2014-08-13 | 西安电子科技大学 | Hyperspectral image regression prediction compression method based on off-line training |
| CN104270640A (en) * | 2014-09-09 | 2015-01-07 | 西安电子科技大学 | Lossless spectrum image compression method based on support vector regression |
| CN105374054A (en) * | 2015-11-17 | 2016-03-02 | 重庆邮电大学 | Hyperspectral image compression method based on spatial spectrum characteristics |
-
2017
- 2017-12-25 CN CN201711421608.1A patent/CN108053455B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103020912A (en) * | 2012-12-18 | 2013-04-03 | 武汉大学 | Remote sensing image restoration method combining wave-band clustering with sparse representation |
| CN103985096A (en) * | 2014-05-18 | 2014-08-13 | 西安电子科技大学 | Hyperspectral image regression prediction compression method based on off-line training |
| CN104270640A (en) * | 2014-09-09 | 2015-01-07 | 西安电子科技大学 | Lossless spectrum image compression method based on support vector regression |
| CN105374054A (en) * | 2015-11-17 | 2016-03-02 | 重庆邮电大学 | Hyperspectral image compression method based on spatial spectrum characteristics |
Non-Patent Citations (3)
| Title |
|---|
| JAMO MIELIKAINEN, ET AL.: "Lossless Hyperspectral Image Compression via Linear Prediction", 《HYPERSPECTRAL DATA COMPRESSION》 * |
| JARNO S. MIELIKAINEN ET AL.: "Lossless hyperspectral image compression via linear prediction", 《SPIE PROCEEDINGS VOL. 4725:ALGORITHMS AND TECHNOLOGIES FOR MULTISPECTRAL, HYPERSPECTRAL, AND ULTRASPECTRAL IMAGERY VIII》 * |
| 粘永健 等: "基于聚类的高光谱图像无损压缩", 《电子与信息学报》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114119787A (en) * | 2021-11-30 | 2022-03-01 | 哈尔滨工业大学 | Hyper-spectral image prediction compression method based on orthogonal matching pursuit |
| CN114119787B (en) * | 2021-11-30 | 2024-04-12 | 哈尔滨工业大学 | Hyper-spectral image prediction compression method based on orthogonal matching pursuit |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108053455B (en) | 2020-06-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Martinez et al. | Permute, quantize, and fine-tune: Efficient compression of neural networks | |
| US8295342B2 (en) | Method and system for efficient video compression with low-complexity encoder | |
| US11412225B2 (en) | Method and apparatus for image processing using context-adaptive entropy model | |
| Rehman et al. | Image compression: A survey | |
| CN102905137B (en) | The quick difference vector of ultraphotic spectrum signal quantizes compaction coding method | |
| Li et al. | Compressing volumetric radiance fields to 1 mb | |
| CN114449276B (en) | Super prior side information compensation image compression method based on learning | |
| CN102300095B (en) | Fast compression coding method for hyperspectral signal and image compression method | |
| Gu et al. | Compression of human motion capture data using motion pattern indexing | |
| Chang et al. | Thousand to one: Semantic prior modeling for conceptual coding | |
| Zhang et al. | Lvqac: Lattice vector quantization coupled with spatially adaptive companding for efficient learned image compression | |
| Rhee et al. | LC-FDNet: Learned lossless image compression with frequency decomposition network | |
| CN116939226A (en) | Low-code-rate image compression-oriented generated residual error repairing method and device | |
| Hou et al. | Scalable and compact representation for motion capture data using tensor decomposition | |
| CN112702600B (en) | Image coding and decoding neural network layered fixed-point method | |
| CN108053455A (en) | A kind of vector quantization method for compressing high spectrum image based on linear prediction decorrelation | |
| CN107231556B (en) | Image cloud storage device | |
| Subramaniam et al. | Low-complexity source coding using Gaussian mixture models, lattice vector quantization, and recursive coding with application to speech spectrum quantization | |
| Jeong et al. | An overhead-free region-based JPEG framework for task-driven image compression | |
| JP2016535382A (en) | Method and apparatus for constructing an original image estimate from a low quality version and epitome of the original image | |
| Hou et al. | Low-latency compression of mocap data using learned spatial decorrelation transform | |
| EP2801952B1 (en) | Method and device for compression of vertex data in three-dimensional image data | |
| CN111897995A (en) | Video feature extraction method and video quantization method applying same | |
| CN103442236A (en) | Remote sensing signal compressed encoding method of multilevel and fractal dimension vector quantization | |
| JP4871246B2 (en) | Vector quantization method and apparatus, their program, and computer-readable recording medium recording the same |
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 |