CN107133992A - Based on spectrum correlation adaptive grouping high spectrum image distributed associating compression method - Google Patents
Based on spectrum correlation adaptive grouping high spectrum image distributed associating compression method Download PDFInfo
- Publication number
- CN107133992A CN107133992A CN201710247646.3A CN201710247646A CN107133992A CN 107133992 A CN107133992 A CN 107133992A CN 201710247646 A CN201710247646 A CN 201710247646A CN 107133992 A CN107133992 A CN 107133992A
- Authority
- CN
- China
- Prior art keywords
- wave band
- image
- mrow
- band
- msub
- 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
- G06T9/007—Transform coding, e.g. discrete cosine transform
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
- G06T2207/10036—Multispectral image; Hyperspectral image
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Discrete Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Compression, Expansion, Code Conversion, And Decoders (AREA)
Abstract
Spectrum correlation adaptive grouping high spectrum image distributed associating compression method is based on the present invention relates to one kind, adaptive grouping is carried out according to the strong and weak of spectrum correlation between each wave band of high spectrum image；It is non-reference wave band to be selected in every group in each wave band with its all band coherence most strong wave band to refer to its all band in wave band, group；Difference operation is done with reference to wave band and non-reference wave band and obtains more sparse residual image, to doing Huffman lossless coding with reference to band image；Section is decoded, distributed compression coding is done to residual image；To residual image combined reconstruction, and then non-reference band image is reconstructed, the reference band image and non-reference band image of reconstruct are combined and obtain whole high-spectral data.The strong wave band of correlation is assigned to one group by the present invention, improves accuracy, and obtaining more preferable effect for the compression of subsequent distribution formula provides guarantee, carries out combined coding using distributed compression perception algorithm, reduces the amount of calculation of algorithm, can obtain good compression effectiveness.
Description
Technical field
The invention belongs to Image Compression field, and in particular to one kind is based on spectrum correlation adaptive grouping high-spectrum
As distributed associating compression method.
Background technology
High-spectrum seem while the set of multiple band images composition comprising spatial information and spectral information, by
Many fields are applied to, such as agricultural, military, geological prospecting and environmental monitoring.However as spatial resolution and spectrum point
The continuous improvement of resolution, brings the data of magnanimity.The data of these magnanimity to high spectrum image storage, transmit and apply band
Carry out huge challenge, therefore how efficiently to realize high-spectral data compression just into urgent problem to be solved.
High spectrum image has more redundancies, compared with traditional remote sensing images and two dimensional image, traditional images
Spatial redundancy is only existed, two kinds of redundancies are primarily present for high spectrum image -- redundancy and spatial redundancy spectrum.Therefore to height
The compression of spectroscopic data is intended to remove between spectrum between the spectrum that exists redundancy and each spectrum spatial redundancy of itself.
At present, Hyperspectral image compression algorithm is essentially divided into three classes：Compression method based on prediction, based on vector quantization
The compression method of technology and the compression method based on conversion.Compression method based on prediction is mainly used in Lossless Compression, compression
By larger limitation, still have high requirements to transmission bandwidth, be unfavorable for the real-time Transmission of data；Based on vector quantization technology
Compression method, its algorithm complex is too high, and amount of calculation is exponentially increased with the increase of vector dimension；Compression side based on conversion
Method utilizes abundant not enough to the characteristic of high spectrum image.
The content of the invention
For compression algorithm in the prior art to the characteristic of high spectrum image using not enough fully so that compressibility is limited etc.
Deficiency, the problem to be solved in the present invention is to provide a kind of based on spectrum correlation adaptive grouping high spectrum image distributed associating pressure
Compression method, to carrying out difference operation with reference to wave band and non-reference wave band, improves openness, distributed associating is carried out to residual image
Coding, obtains more preferable compression effectiveness.
In order to solve the above technical problems, the technical solution adopted by the present invention is：
One kind of the invention is based on spectrum correlation adaptive grouping high spectrum image distributed associating compression method, including following
Step：
1) correlation analysis is carried out to each wave band of high spectrum image, carried out according to the power of spectrum correlation between each wave band adaptive
It should be grouped；
2) the correlation power to wave band in every group is compared, by least square method select in each wave band with other ripples
Section coherence most strong wave band is non-reference wave band as its all band in wave band, group is referred to；
3) difference operation is done by reference to wave band and non-reference wave band and obtains more sparse residual image, to referring to wave band figure
As doing Huffman lossless coding, distributed compression coding is done to residual image；
4) image with reference to wave band is recovered using Hafman decoding, is perceived with reference to the image of wave band as distributed compression
Priori conditions, and then to residual image carry out combined reconstruction, the reference band image of reconstruct is added with residual image and obtained
Non-reference band image, the reference band image and non-reference band image of reconstruct are combined and obtain whole high-spectral data.
Adaptive grouping is carried out according to the strong and weak of spectrum correlation between each wave band, comprised the following steps：
11) each wave band and the correlation coefficient r of its adjacent bandk(m)：
Wherein, rk(m) be kth wave band vector xkIn m elements：It is k-th of band image
The average value of middle all pixels, S is number of pixels；
If 12) rk(m) ＞ R, R are a threshold value set in advance, then wave band k and (k+1) are assigned into same subset n
In, otherwise k is assigned in subset (n+1)；
13) number of each subset medium wave band is calculated；
If only one of which wave band in subset, Huffman lossless compression-encoding is used to this wave band, otherwise using packet
Wave band in subset is divided into some groups, the wave band number being each grouped is set as L, then：
If the number of the wave band in subset is less than L, whole subset is regard as a group；
Otherwise L adjacent wave bands are assigned in a group, when the number of last group is less than L, by last
Organize and penultimate group merges into one group.
The reference wave band is determined by following steps：
Z=[X1,X2,...Xi,...Xm] represent the matrix that all wave bands are constituted in any packet, wherein XiFor each wave band
Matrix form, m is the quantity of this group of wave band, L≤m ＜ 2L, then the optimal reference wave band λ in the packet can be following excellent by solving
Change problem is obtained：
In formula,For the wave band number in the packet,Rz(i, j) represents the intersegmental phase of the packet medium wave
Relation number, the wave band obtained under least square method criterion by solving λ is then to have maximum phase with its all band in the packet
The wave band of closing property, that is, elect optimal reference wave band as, i, and j is the coordinate of pixel.
Residual image corresponding to non-reference wave band carries out distributed compression perceptual coding, comprises the following steps：
Optimal reference wave band is determined according to the step of determination optimal reference wave band noted earlier, it is assumed herein that X1For reference wave
Section, i.e., the public non-sparse part of each signal, other non-reference wave bands and X1Difference operation is done, corresponding residual plot is obtained,
I.e.：
In formula,Each column vector represent that high spectrum image does the residual plot matrix after difference operation, i.e., each signal
Distinctive sparse part, is observed by compressed sensing to residual image, the data after being encoded：
Wherein Φ be calculation matrix, Z be all measurement sets of each residual image into matrix, ziRepresent residual image correspondence
Measured value column vector, m is the quantity of this group of wave band.
Beneficial effects of the present invention：
1. the EO-1 hyperion compression method that the present invention is provided, according to the adaptive grouping of spectrum correlation, by the strong ripple of correlation
Section assigns to one group, improves the accuracy of packet, and obtaining more preferable effect to the compression of subsequent distribution formula provides guarantee.
2. non-reference band image is carried out difference operation pretreatment by the inventive method, obtained residual image has higher
It is openness, using distributed compression perception algorithm carry out combined coding, reduce high-spectral data dimension, reduce algorithm
Amount of calculation, and good compression effectiveness can be obtained.
Brief description of the drawings
Fig. 1 is the curve of the correlation of high spectrum image Terrain adjacent spectral bands；
Fig. 2 is the curve of the correlation of high spectrum image cuprite adjacent spectral bands；
Fig. 3 is the flow chart of the inventive method；
Fig. 4 is the schematic diagram of high spectrum image adaptive grouping；
Fig. 5 is the distributed compression schematic diagram of wave band in high spectrum image group；
The curve that Fig. 6 changes for the average peak signal to noise ratio of four kinds of algorithms of Terrain images with sample rate；
The curve that Fig. 7 changes for the average peak signal to noise ratio of four kinds of algorithms of Cuprite images with sample rate；
Fig. 8 (a) is the different-waveband of Terrain images respectively by classical CS algorithms in average sample rate SR=40%
Reconstruction image；
Fig. 8 (b) composes combined coding algorithm in average sample rate SR=by sky respectively for the different-waveband of Terrain images
Reconstruction image when 40%；
Fig. 8 (c) is the different-waveband of Terrain images respectively by AG_JSM-3 algorithms in average sample rate SR=40%
Reconstruction image；
Fig. 9 (a) is the different-waveband of Cuprite images respectively by classical CS algorithms in average sample rate SR=40%
Reconstruction image.
Fig. 9 (b) composes combined coding algorithm in average sample rate SR=by sky respectively for the different-waveband of Cuprite images
Reconstruction image when 40%.
Fig. 9 (c) is the different-waveband of Cuprite images respectively by AG_JSM-3 algorithms in average sample rate SR=40%
Reconstruction image.
Embodiment
The embodiment of the present invention is further elaborated with reference to Figure of description.
The compression method of high spectrum image of the invention based on spectrum correlation adaptive grouping is as shown in Figure 3.Including following
Step：
1) correlation analysis is carried out to each wave band of high spectrum image, carried out according to the power of spectrum correlation between each wave band adaptive
It should be grouped；
2) the correlation power to wave band in every group is compared, by least square method select in each wave band with other ripples
Section coherence most strong wave band is non-reference wave band as its all band in wave band, group is referred to；
3) difference operation is done by reference to wave band and non-reference wave band and obtains more sparse residual image, to referring to wave band figure
As doing Huffman lossless coding, distributed compression coding is done to residual image；
4) image with reference to wave band is recovered using Hafman decoding, is perceived with reference to the image of wave band as distributed compression
Priori conditions, and then to residual image carry out combined reconstruction, the reference band image of reconstruct is added with residual image and obtained
Non-reference band image, the reference band image and non-reference band image of reconstruct are combined and obtain whole high-spectral data.
Step 1) in, multiple band images of high spectrum image are grouped according to the power of spectrum correlation；
High spectrum image spectrum correlation is illustrated first.Picture of the high spectrum image different spectral bands in the same space position
The similitude of element, the reason for producing this spectrum pixel similarity has at following 2 points：First, the pixel of each bands of a spectrum of spectrum picture is
The reflection intensity values of atural object, the reflectivity of the atural object of adjacent spectral band is similar.2nd, due to involved by the image of different spectral bands
Ground target is identical, and they have identical space structure.In order to analyze the correlation between wave band, cross-correlation function is introduced
Concept.Cross-correlation is defined as：
H (k, l)=∫ ∫ (f (x+k, y+l)-uf)(g(x,y)-ug)dxdy
Wherein, f (x, y) and g (x, y) is the gray value function of different spectral bands image, ufAnd ugFor being averaged for gradation of image
Value, referred to as h (k, l) is f (x, y) and g (x, y) cross-correlation function, and x, y is the coordinate of pixel, and l, k is pixel point coordinates
Offset.
Above formula is normalized and sliding-model control, had：
Wherein,The position that l, k represent ranks respectively becomes
Change value, M, N are the size of the matrix of image.
Coefficient correlation between different spectral bands is：
If bands of a spectrum i image is fi(x, y), bands of a spectrum i+1 image is fi+1(x, y), defines bands of a spectrum i adjacent spectral band
Coefficient correlation is：
Spectrum correlation between Cuprite and Terrain adjacent bands can be calculated according to the formula.
Adjacent band has very strong correlation it can be seen from Fig. 1 and Fig. 2, and this is the weight of distributed compression aware application
Premise is wanted, but the correlation of high spectrum image Terrain adjacent spectrals band is not very stable, adjacent wave as can be seen from Figure 1
The strong and weak jump of section correlation is very big, and this compression recovery effects for allowing for distributed compression perception cannot ensure well,
Based on this, the invention proposes to be grouped each wave band of EO-1 hyperion using the method for spectrum correlation adaptive grouping, Ran Hou
Carry out distributed compression.It is specific as follows：
Step 11) calculate correlation coefficient r between all adjacent bandsk(m), maximizing rmaxWith average value rV.Set
Threshold value R=rV* δ (wherein 1 ＜ δ ＜ rmax/rV)；
R in formulak(m) be kth wave band vector xkIn m elements：It is institute in kth band image
There is the average value of pixel, S is the number of pixel；
Step 12) if rk(m) ＞ R, then assign to wave band k and (k+1) in same subset n, k otherwise assigned into subset
(n+1) in.
Step 13) calculate the number of each subset medium wave band.The wave band that antithetical phrase is concentrated is grouped, as shown in Figure 4；Bloom
Each wave band of spectrogram picture is divided into some groups of such as N groups according to the power of spectrum correlation, and every group of wave band number is also different.Pass through scanning
The situation of every group of medium wave band number, carries out different processing.If only one of which wave band in subset, Hough is used to this wave band
Graceful lossless compression-encoding, is otherwise divided into some groups using packet by the wave band in subset, set the wave band number that is each grouped as
L, then：
If the number of the wave band in subset is less than L, whole subset is regard as a group；Otherwise by L adjacent wave bands
Assign in a group, when the number of last group is less than L, last group and penultimate group are merged into one
Group.
Step 2) in, the compression effectiveness that distributed compression is perceived depends on reference signal selection, reference signal and non-reference
The coherence of signal is stronger, and it is better that effect is rebuild in compression.Based on this requirement, correlation analysis is carried out to each group of wave band,
Select in each group with remaining wave band correlation most strong wave band as referring to wave band., should for the packet of only one of which wave band
Wave band is the best band of the group.For the packet Z=[X of a unnecessary wave band1,X2,...Xi,...Xm] represent packet n
In all wave bands composition matrix, wherein XiFor each multiband matrix form, m is the quantity of this group of wave band, wherein L≤m ＜ 2L.
Then packet n optimal reference wave band λ can be obtained by solving following optimization problem：
In formulaRz(i, j) represents the intersegmental coefficient correlation of packet n medium waves, is obtained under criterion of least squares
The wave band obtained is then the wave band with all band of its in packet with maximum correlation, i.e. optimal reference wave band.
The optimal reference wave band of each packet can be obtained successively according to above method.
Step 3) in, according to distributed compression perception theory, distributed compression is carried out to residual image wave band, such as Fig. 5 institutes
Show；
The spatial coherence of high spectrum image is weaker, and spectrum correlation is but very strong, illustrates between each spectral coverage of high spectrum image
Similitude is very big.Applied compression perception theory of the CS algorithms only to high spectrum image different-waveband image independence, only eliminates figure
The spatial redundancy information of picture, redundancy between the spectrum of image is not considered.And empty spectrum joint (S-s-CS) encryption algorithm, without logarithm
According to sparse processing is carried out, spectrum correlation is not also removed completely.
Therefore, the difficult point of Compression of hyperspectral images is how to remove Spectral correlation.JSM-3 is perceived in distributed compression
In model, all signals are all made up of a public non-sparse part and the distinctive sparse part of each signal.Although high
Each wave band of spectrum picture is non-sparse image, but has very strong correlation between image, so doing difference between wave band
The residual image of operation generation is sparse graph picture, therefore the JSM-3 models that distributed compression can be perceived are applied to high-spectrum
As in compression, sampling is compressed to it with lower sample rate.
The shortcoming existed for CS algorithms with empty spectrum combined coding algorithm, the present invention is to high spectrum image applied compression sense
Know before observation measurement, redundancy between this strong spectrum is first removed using the technology of linear prediction, in traditional high spectrum image
In pre- measured compressed, the method for compressing high spectrum image based on pre- assize is proposed that this method is first obtained a certain by Menomon etc. earliest
The residual image of individual wave band, then finds a spanning tree so that macro-forecast residual error using the method for minimum spanning tree in figure
It is minimum.Above-mentioned thought is used for reference, high spectrum image is converted to by JSM-3 models by prediction residual method.Use matrix X=
[X1,...,Xi,...,Xk] represent one group of high spectrum image data, wherein k be high spectrum image wave band number, each row Xi
The i-th width band image is represented according to the data after the processing of row one-dimensional, its data amount check is the pixel count N of band image, i.e. X is
N × k matrix.Assuming that the group has k wave band, what is obtained by criterion of least squares most joins by force with its all band correlation in group
Wave band is examined for Xi, wherein 0 ＜ i ＜ k.Assuming that X1To refer to wave band, i.e., the public non-sparse part of each signal will refer to wave band
X1Difference operation is done with non-reference wave band, i.e.,：
Each column vector represent that high spectrum image does the residual plot matrix after difference operation, i.e., each signal is distinctive
Sparse part.All packets are all done into difference operation according to above method, more sparse residual image data is obtained.
By high spectrum image after difference processing, data become more sparse, and sample rate greatly reduces.Study table
Bright, if calculation matrix is incoherent with sparse transformation matrix, the two product meets limited equidistant standard in very maximum probability
Then；Gaussian random matrix and Bernoulli Jacob's random matrix can ensure all irrelevant with most sparse transformation matrix；However,
When handling digital picture with compressed sensing, gaussian random matrix or Bernoulli Jacob's random matrix are not very applicable, because a width
The dimension of the column vector of image is very big, and the calculation matrix of gaussian random matrix or Bernoulli Jacob's random matrix construction is even more
It is very huge, and the matrix data type of generation is floating type, and this, which results in measurement process, needs very big memory headroom
And amount of calculation, cost height, it is unfavorable for the realization of hardware device；
Present embodiment will be applied to image with one kind and be easy to hard-wired calculation matrix --- part Hadamard
Calculation matrix, hadamard matrix be by+1 and -1 element constitute and meet AA'=nE (here A' be A transposition, E is unit
Square formation, n is order of matrix number), the construction of hadamard matrix only needs to plus and minus calculation, and amount of calculation is small, and measurement process need not be accounted for
With very big memory headroom.Its building method is：The hadamard matrix of N × N size is firstly generated, it is then random from this
M row vectors are chosen in hadamard matrix, the calculation matrix that a size is M × N is constituted.Because hadamard matrix is orthogonal moment
Battle array, therefrom taking the part hadamard matrix of the M × N sizes obtained after M rows still has stronger irrelevant row and part just
The property handed over, so compared with other certainty matrixes, the measurement number required for the calculation matrix Exact Reconstruction is less, that is to say, that
Under same measurement number, the reconstruction effect of part hadamard matrix is relatively good.
Each residual image independence by a M × N (M ＜ N) part Hadamard calculation matrix Φ, by it from a N-dimensional
Spatial sampling is to M dimension spaces, to the i-th wave band residual image, wherein 0 ＜ i ＜ k, sampling observation process is as follows：
The observation of obtained residual image is：
Wherein Φ be calculation matrix (Φ line number will be far smaller than columns), Z be each all measurement sets of difference wave band into
Matrix, ZiFor the observation of a width residual image, wherein 0 ＜ i ＜ k, k is the quantity of residual image.After observation, in Z
Data volume be far smaller than residual imageData volume.For the public non-sparse part X of signal1, using traditional Hough
Graceful lossless compression-encoding, to ensure that later reconstitution goes out high-quality original image.
Step 4) in, different coding process is decoded accordingly in decoding end, combined reconstruction goes out high spectrum image；
The image X with reference to wave band is recovered first with Hafman decoding1, it regard the view data with reference to wave band as distribution
The priori conditions of formula compressed sensing combined reconstruction, residual image data is gone out with TVAL3 restructing algorithm combined reconstruction:
In formula,For the residual error data of the i-th row reconstruct, the reference band image of reconstruct is added with residual image and obtains non-
With reference to band image
The reference band image reconstructed in all groups and non-reference band image are combined and obtain whole high spectrum image number
According to.
Illustrate the superiority of present embodiment with reference to experimental data and experimental result：
From AVIRIS high spectrum image the sequence Cs uprite and Terrain in JPL laboratories the first scape image, AVIRI
High spectrum image is made up of 224 wave bands, spectral region covering visible light to near-infrared (400nm~2500nm).Each pixel
Stored with 16 than peculiar symbol integer.The size of each group of high spectrum image be 512 × 614 × 224 × 16bit (it is long × wide ×
Wave band × bit-depth).For the ease of calculating, 64 × 64 region image data collection experimental results are only taken.Calculation matrix Φ
From part hadamard matrix.The definition sampling rate of SR=M/N × 100%, M is observation number, and N is signal length.Rebuild and calculate
Method selects TVAL3 algorithms.
In order to verify the performance of this programme, 4 have been done under identical calculating platform (Intel monokaryon 2.70GHZ/2G internal memories)
Group experiment：1) classical CS algorithms；2) S-s-CS (sky-spectrum combined coding) algorithm；3) AG-JSM-3 (adaptive grouping) algorithm.
Table 1 is believed giving average peak of four kinds of algorithms under different sample rates in Cuprite images Terrain images and table 2
Make an uproar than (PSNR).
Average peak signal to noise ratio (PSNR) of table 1 Terrain images, the four kinds of algorithms under different sample rates
Terrain image sequences, compared with classical CS algorithms, AG-JSM-3 algorithms when sample rate SR is higher
Average PSNR obtains nearly 12dB lifting, and can more embody the advantage of AG-JSM-3 algorithms when sample rate is low, is adopting
During sample rate SR=20%, AG-JSM-3 algorithms improve nearly 15dB than the average PSNR of classical CS algorithms.And calculated with S-s-CS
Method is compared, and PSNR averagely improves 8.53dB, and in sample rate SR=10%, PSNR improves 8.66dB, is effectively improved
Recover the quality of image, because the residual image of high spectrum image is openness very strong, it is sufficient with seldom measured value
Access the reconstructed image of certain mass.
Average peak signal to noise ratio (PSNR) of table 2 Cuprite images, the four kinds of algorithms under different sample rates
Cuprite image sequences, compared with CS algorithms, AG-JSM-3 algorithms being averaged when sample rate SR is 60%
PSNR obtains nearly 10.71db lifting, and can more embody the advantage of AG-JSM-3 algorithms when sample rate is low, in sampling
During rate SR=10%, AG-JSM-3 algorithms improve nearly 14.69dB than the average PSNR of classical CS algorithms, combine with sky-spectrum
Encryption algorithm is compared and also improves 10.01dB.It is effectively improved the quality for recovering image.
In order to reflect influence of the sample rate to algorithm, Fig. 6 and Fig. 7 depict Terrain, Cuprite image sequence respectively
Tri- kinds of algorithms of CS, S-s-CS, AG-JSM-3 the curve that changes with sample rate of average peak signal to noise ratio.
The average PSNR of recovery from Fig. 6 and Fig. 7 image sequences drawn can be seen that the compression effectiveness of AG-JSM-3 algorithms
On be substantially better than traditional compression method.
In order to more intuitively compare the quality for recovering image, Fig. 8 (a)~8 (c), Fig. 9 (a)~9 (c) give 2 width images
The reconstruction image of the different-waveband of sequence respectively by three kinds of algorithms in average sample rate SR=40%.Seen by visual effect subjectivity
Examine image Quality of recovery.
For Terrain images, it can be seen that classics CS algorithm patterns picture recovers relatively to obscure, detailed information is not
It is prominent, than improving 4.23dB compared with sky-spectrum combined coding algorithm, but it is extensive in the less wave band of high spectrum image spectrum correlation
Multiple effect is unsatisfactory, for example the recovery effects of the 87th wave band substantially than classical CS compression algorithms and with sky-spectrum combined coding
Algorithm is far short of what is expected, and PSNR only has 1.69dB, and image is not recovered.And this paper AG-JSM-3 algorithms, it improves tradition
The defect of algorithm, lossless coding is used by the less wave band of correlation, and all wave bands high-quality can recover, in such as Fig. 8 (c)
87th wave band PSNR is inf, is greatly improved.
Fig. 9 (a)~9 (c) is that Cuprite images are carried for the PSNR of the 36th band image compared to classical CS compression algorithms
High more, PSNR reaches more than 39dB, can substantially be contrasted by visual image, and AG-JSM-3 algorithms compress than classical CS to be calculated
Method and sky-spectrum combined coding compression algorithm efficiency improve a lot.
It can be seen that from Fig. 6 and Fig. 7 and Fig. 8 (a)~8 (c), Fig. 9 (a)~9 (c), for Cuprite, Terrain image
Sequence, experiment shows that this paper AG-JSM-3 algorithms are effectively utilized stronger spectrum correlation, improves compression performance well, carries
The high reconstruction quality of compression image, is enhanced, effect is substantially better than other three kinds of algorithms, tests in compression efficiency
The feasibility and superiority of algorithm are demonstrate,proved.
Claims (4)
1. one kind be based on spectrum correlation adaptive grouping high spectrum image distributed associating compression method, it is characterised in that including with
Lower step：
1) correlation analysis is carried out to each wave band of high spectrum image, adaptively divided according to the power of spectrum correlation between each wave band
Group；
2) the correlation power to wave band in every group is compared, by least square method select in each wave band with its all band phase
Dryness most strong wave band is non-reference wave band as its all band in wave band, group is referred to；
3) difference operation is done by reference to wave band and non-reference wave band and obtains more sparse residual image, to being done with reference to band image
Huffman lossless coding, distributed compression coding is done to residual image；
4) image with reference to wave band is recovered using Hafman decoding, with reference to the elder generation that is perceived as distributed compression of image of wave band
Condition is tested, and then combined reconstruction is carried out to residual image, the reference band image of reconstruct is added with residual image and obtains non-ginseng
Band image is examined, the reference band image and non-reference band image of reconstruct are combined and obtain whole high-spectral data.
2. as described in claim 1 based on spectrum correlation adaptive grouping high spectrum image distributed associating compression method, its
It is characterised by carrying out adaptive grouping according to the strong and weak of spectrum correlation between each wave band, comprises the following steps：
11) each wave band and the correlation coefficient r of its adjacent bandk(m)：
<mrow>
<msub>
<mi>r</mi>
<mi>k</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>m</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mrow>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>m</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>S</mi>
</munderover>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>k</mi>
</msub>
<mo>(</mo>
<mi>m</mi>
<mo>)</mo>
<mo>-</mo>
<msub>
<mover>
<mi>x</mi>
<mo>&OverBar;</mo>
</mover>
<mi>k</mi>
</msub>
<mo>)</mo>
</mrow>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mrow>
<mi>k</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>(</mo>
<mi>m</mi>
<mo>)</mo>
<mo>-</mo>
<msub>
<mover>
<mi>x</mi>
<mo>&OverBar;</mo>
</mover>
<mrow>
<mi>k</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>)</mo>
</mrow>
</mrow>
<msqrt>
<mrow>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>m</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>S</mi>
</munderover>
<msup>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>k</mi>
</msub>
<mo>(</mo>
<mi>m</mi>
<mo>)</mo>
<mo>-</mo>
<msub>
<mover>
<mi>x</mi>
<mo>&OverBar;</mo>
</mover>
<mi>k</mi>
</msub>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>m</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>S</mi>
</munderover>
<msup>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mrow>
<mi>k</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>(</mo>
<mi>m</mi>
<mo>)</mo>
<mo>-</mo>
<msub>
<mover>
<mi>x</mi>
<mo>&OverBar;</mo>
</mover>
<mrow>
<mi>k</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
</msqrt>
</mfrac>
</mrow>
Wherein, rk(m) be kth wave band vector xkIn m elements：It is institute in k-th of band image
There is the average value of pixel, S is number of pixels；
If 12) rk(m) ＞ R, R are a threshold value set in advance, then assign to wave band k and (k+1) in same subset n, no
Then k is assigned in subset (n+1)；
13) number of each subset medium wave band is calculated；
If only one of which wave band in subset, Huffman lossless compression-encoding is used to this wave band, packet is otherwise used by son
The wave band of concentration is divided into some groups, sets the wave band number being each grouped as L, then：
If the number of the wave band in subset is less than L, whole subset is regard as a group；
Otherwise L adjacent wave bands are assigned in a group, when the number of last group is less than L, by last group
One group is merged into penultimate group.
3. as described in claim 1 based on spectrum correlation adaptive grouping high spectrum image distributed associating compression method, its
It is characterised by：The reference wave band is determined by following steps：
Z=[X1,X2,...Xi,...Xm] represent the matrix that all wave bands are constituted in any packet, wherein XiFor each wave band arrange to
Moment matrix form, m is the quantity of this group of wave band, L≤m ＜ 2L, then the optimal reference wave band λ in the packet can be following by solving
Optimization problem is obtained：
In formula,For the wave band number in the packet,Rz(i, j) represents the intersegmental phase relation of the packet medium wave
Number, the wave band obtained under least square method criterion by solving λ is then to have maximum correlation with its all band in the packet
Wave band, that is, elect optimal reference wave band as, i, j is the coordinate of pixel.
4. as described in claim 1 based on spectrum correlation adaptive grouping high spectrum image distributed associating compression method, its
It is characterised by, residual image corresponding to non-reference wave band carries out distributed compression perceptual coding, comprises the following steps：
Determine optimal reference wave band, it is assumed that X1To refer to wave band, i.e., the public non-sparse part of each signal, other non-reference ripples
Section and X1Difference operation is done, corresponding residual plot is obtained, i.e.,：
In formula,Each column vector represent that high spectrum image does the residual plot matrix after difference operation, i.e., each signal is peculiar
Sparse part, residual image is observed by compressed sensing, the data after being encoded：
Wherein Φ be calculation matrix, Z be all measurement sets of each residual image into matrix, ziRepresent that residual image is corresponding to survey
Value column vector, m is the quantity of this group of wave band.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710247646.3A CN107133992B (en) | 2017-04-17 | 2017-04-17 | Based on spectrum correlation adaptive grouping high spectrum image distributed associating compression method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710247646.3A CN107133992B (en) | 2017-04-17 | 2017-04-17 | Based on spectrum correlation adaptive grouping high spectrum image distributed associating compression method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107133992A true CN107133992A (en) | 2017-09-05 |
| CN107133992B CN107133992B (en) | 2019-07-12 |
Family
ID=59715024
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710247646.3A Expired - Fee Related CN107133992B (en) | 2017-04-17 | 2017-04-17 | Based on spectrum correlation adaptive grouping high spectrum image distributed associating compression method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107133992B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109272561A (en) * | 2018-10-09 | 2019-01-25 | 西安航空学院 | Based on the empty spectrum joint Compression of hyperspectral images sensing reconstructing methods for assuming prediction more |
| CN109682476A (en) * | 2019-02-01 | 2019-04-26 | 北京理工大学 | A method of compression high light spectrum image-forming is carried out using adaptive coding aperture |
| CN110390699A (en) * | 2019-07-17 | 2019-10-29 | 中国人民解放军陆军军医大学 | A kind of compressed sensing based high spectrum image distributed compression method and system |
| CN110988899A (en) * | 2019-12-09 | 2020-04-10 | Oppo广东移动通信有限公司 | Method for removing interference signal, depth detection assembly and electronic device |
| CN111385582A (en) * | 2020-04-13 | 2020-07-07 | 辽宁工程技术大学 | Hyperspectral remote sensing image lossless compression algorithm based on spectrum-space joint prediction |
| CN114002165A (en) * | 2021-10-29 | 2022-02-01 | 中国科学院新疆生态与地理研究所 | Copper element abundance prediction method based on copper element spectral index inversion |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101720043A (en) * | 2009-11-20 | 2010-06-02 | 北京工业大学 | Imaging spectrum image compression method based on multi-mode prediction |
| CN101883274A (en) * | 2009-05-08 | 2010-11-10 | 中国科学院沈阳自动化研究所 | Spatial-spectral associated prediction-based hyperspectral image lossless compression method |
| CN101893552A (en) * | 2010-07-06 | 2010-11-24 | 西安电子科技大学 | Hyperspectral imager and imaging method based on compressive sensing |
| CN103903261A (en) * | 2014-03-24 | 2014-07-02 | 西安电子科技大学 | Spectrum image processing method based on partition compressed sensing |
-
2017
- 2017-04-17 CN CN201710247646.3A patent/CN107133992B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101883274A (en) * | 2009-05-08 | 2010-11-10 | 中国科学院沈阳自动化研究所 | Spatial-spectral associated prediction-based hyperspectral image lossless compression method |
| CN101720043A (en) * | 2009-11-20 | 2010-06-02 | 北京工业大学 | Imaging spectrum image compression method based on multi-mode prediction |
| CN101893552A (en) * | 2010-07-06 | 2010-11-24 | 西安电子科技大学 | Hyperspectral imager and imaging method based on compressive sensing |
| CN103903261A (en) * | 2014-03-24 | 2014-07-02 | 西安电子科技大学 | Spectrum image processing method based on partition compressed sensing |
Non-Patent Citations (4)
| Title |
|---|
| NASIR D. MEMON 等: "Lossless Compression of Multispectral Image Data", 《IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING》 * |
| PARUL SHAH 等: "Efficient hierarchical fusion using adaptive grouping techniques for visualization of hyperspectral images", 《2011 8TH INTERNATIONAL CONFERENCE ON INFORMATION, COMMUNICATIONS & SIGNAL PROCESSING》 * |
| 袁晓玲 等: "一种高光谱图像分布式压缩感知重构方法", 《电子设计工程》 * |
| 陈善学 等: "基于自适应波段聚类PCA的高光谱图像压缩", 《科学技术与工程》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109272561A (en) * | 2018-10-09 | 2019-01-25 | 西安航空学院 | Based on the empty spectrum joint Compression of hyperspectral images sensing reconstructing methods for assuming prediction more |
| CN109272561B (en) * | 2018-10-09 | 2022-09-30 | 西安航空学院 | Hyperspectral image compressed sensing reconstruction method based on space-spectrum joint multi-hypothesis prediction |
| CN109682476A (en) * | 2019-02-01 | 2019-04-26 | 北京理工大学 | A method of compression high light spectrum image-forming is carried out using adaptive coding aperture |
| CN109682476B (en) * | 2019-02-01 | 2020-06-19 | 北京理工大学 | Method for performing compressed hyperspectral imaging by using self-adaptive coding aperture |
| CN110390699A (en) * | 2019-07-17 | 2019-10-29 | 中国人民解放军陆军军医大学 | A kind of compressed sensing based high spectrum image distributed compression method and system |
| CN110390699B (en) * | 2019-07-17 | 2023-03-24 | 中国人民解放军陆军军医大学 | Hyperspectral image distributed compression method and system based on compressed sensing |
| CN110988899A (en) * | 2019-12-09 | 2020-04-10 | Oppo广东移动通信有限公司 | Method for removing interference signal, depth detection assembly and electronic device |
| CN111385582A (en) * | 2020-04-13 | 2020-07-07 | 辽宁工程技术大学 | Hyperspectral remote sensing image lossless compression algorithm based on spectrum-space joint prediction |
| CN114002165A (en) * | 2021-10-29 | 2022-02-01 | 中国科学院新疆生态与地理研究所 | Copper element abundance prediction method based on copper element spectral index inversion |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107133992B (en) | 2019-07-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107133992A (en) | Based on spectrum correlation adaptive grouping high spectrum image distributed associating compression method | |
| CN107690070B (en) | Based on distributed video compression perceptual system and method without feedback code rate control | |
| AU785112B2 (en) | System and method for encoding/decoding multidimensional data using successive approximation multi-stage vector quantisation | |
| US20140193078A1 (en) | Optimized orthonormal system and method for reducing dimensionality of hyperspectral images | |
| CN101883274B (en) | Spatial-spectral associated prediction-based hyperspectral image lossless compression method | |
| CN108960333A (en) | Lossless compression method for high spectrum image based on deep learning | |
| CN103903261A (en) | Spectrum image processing method based on partition compressed sensing | |
| CN102215385B (en) | Real-time lossless compression method for image | |
| CN112149652A (en) | Space-spectrum joint depth convolution network method for lossy compression of hyperspectral image | |
| CN103763558B (en) | texture image compression method based on similarity | |
| CN113689513A (en) | SAR image compression method based on robust tensor decomposition | |
| CN107124612B (en) | Method for compressing high spectrum image based on distributed compression perception | |
| CN107920250B (en) | Compressed sensing image coding transmission method | |
| CN105825530B (en) | Littoral zone high spectrum image distribution lossy coding and coding/decoding method based on area-of-interest | |
| CN108390871B (en) | Radar data compression method based on autoregressive model frame prediction | |
| CN104683818B (en) | Method for compressing image based on biorthogonal invariant set m ultiwavelet | |
| CN108495134B (en) | Bayer image compression method based on JPEG2000 standard | |
| Cagnazzo et al. | Low-complexity compression of multispectral images based on classified transform coding | |
| CN103761753B (en) | Decompression method based on texture image similarity | |
| Al-Rawi et al. | Image compression using contourlet transform | |
| CN102281443A (en) | Method for processing compressed sensing image based on optimized hierarchical discrete cosine transform (DCT) | |
| CN110930466B (en) | Hyperspectral self-adaptive compressive sensing method for BOIs with arbitrary shape | |
| CN104486631B (en) | A kind of remote sensing image compression method based on human eye vision Yu adaptive scanning | |
| BH et al. | Overview on machine learning in image compression techniques | |
| Kumar et al. | A comparative case study on compression algorithm for remote sensing images |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190712 |