CN113038143A - Hyper-spectral image lossless compression coding system - Google Patents

Abstract

The invention discloses a lossless compression coding system for hyperspectral images. The system utilizes parameterized three-band inter-spectral integer reversible transformation combined with wavelet lifting scheme transformation to remove inter-spectral transformation, removes inter-spectral and spatial redundancy, and adopts SPIHT coding algorithm to realize Lossless compression of hyperspectral images, de-redundancy transforms include: XCJRCT transform and wavelet transform promotion scheme, SPIHT coding algorithm is optimized and analyzed based on time complexity algorithm through parallel operation of encoder and decoder, anti-interference and fault tolerance. The invention solves the problems of low lossless compression efficiency and low compression ratio of the existing hyperspectral images.

Description

A Lossless Compression Coding System for Hyperspectral Images

technical field

The invention relates to the field of hyperspectral images, in particular to a lossless compression coding system for hyperspectral images.

Background technique

Hyperspectral remote sensing images have important application prospects in the fields of agriculture, geography, resources and military. In the field of agricultural engineering, spectral detection sensors are used to obtain timely and accurate information on soil nutrient content in cultivated land, and correlation analysis is used to extract sensitive bands of soil available phosphorus content. And characteristic waveband combination, through the wavelet transform of hyperspectral image data reflectivity, the soil available phosphorus spectral diagnosis model with various sensitive waveband combinations is established. Relying on spectral technology and computer vision technology as research methods, it is of wider practical value to study the automatic detection method of crop environmental ecological factors and analyze the characteristics of spectral and image data.

Inter-spectral transformation is an important and key step in image processing such as multi-spectral images and hyperspectral images. High compression ratio is achieved through inter-spectral transformation. The research on hyperspectral image compression and coding mainly focuses on de-redundancy and coding algorithms. The traditional typical compression algorithm has low lossless compression efficiency and low compression ratio, which is difficult to meet the demand.

SUMMARY OF THE INVENTION

Therefore, the present invention provides a hyperspectral image lossless compression coding system to solve the problems of low lossless compression efficiency and low compression ratio of the existing hyperspectral images.

In order to achieve the above object, the present invention provides the following technical solutions:

The invention discloses a lossless compression coding system for hyperspectral images. The system utilizes parameterized three-band inter-spectral integer reversible transformation combined with wavelet lifting scheme transformation to remove inter-spectral transformation, removes inter-spectral and spatial redundancy, and adopts SPIHT coding algorithm to realize Lossless compression of hyperspectral images, de-redundancy transforms include: XCJRCT transform and wavelet transform promotion scheme, SPIHT coding algorithm is optimized and analyzed based on time complexity algorithm through parallel operation of encoder and decoder, anti-interference and fault tolerance.

Further, the XCJRCT transform realizes the matrix reversible transformation with three-band inter-spectral integers, and performs the optimal transformation solution, and the three-band inter-spectral integer reversible transformation matrix display form XCJRCT transforms into:

Among them, γ and λ are parameters that can be adjusted during transformation, and their numerical values can be selected according to the actual needs of dealing with problems. In ₁ , In ₂ , and In ₃ are input signals, and Out ₁ , Out ₂ , and Out ₃ are output signals. '>>' is the binary left shift symbol.

Further, the wavelet transform promotion scheme includes: splitting, predicting, updating and optimizing promotion;

Splitting, splitting the original signal S _j,k into two mutually disjoint subsets S _j+1,k and d _j+1 , usually first perform Lazy wavelet or Polyphase wavelet transform on the original signal S _j,k , and transform An original signal sequence is divided into an even sequence and an odd sequence, namely split(S _j,k )=(S _j,2k ,S _j,2k+1 )=(S _j+1,k ,d _j+1,k );

For prediction, for the correlation between data, S _j+1,k can be used to predict d _j+1,k , so a prediction operator P that is independent of the structure of the data set can be used, so that d _j+1 ,k=P( S _j+1,k ), use the difference between d _j+1,k and the predicted value P(S _j+1,k ) to replace d _j+1,k , then this difference reflects the approximation degree of the two , if the prediction is reasonable, the difference data set contains less information than the original subset d _j+1,k ;

Update, some overall properties of the coefficient subset S _j+1,k generated by the two steps of splitting and prediction are not consistent with the properties of the original data, and an update process is required to generate a better sub-set through the operator U. The data set S _j+1,k keeps some characteristics of the original data set S _{j,k ,} the definition of S _{j+1,k is S j+1,k} =S _j,2k+1 +U(d _{j +1,k} ), perform the same splitting, prediction and update for the data subset S _j+1,k, then S _j+1,k can be decomposed into d _j+2,k and S _j+2,k , After decomposing J times, the wavelet transform of the original data S _0,k is expressed as {S _J ,d _J ,d _J-1 ,...,d ₁ }, where S _J represents the low-frequency part of the signal, and {d _J ,d _J-1 ,...,d ₁ } is the high frequency part of the signal,

Optimized lifting, applying dual lifting steps and update lifting steps alternately according to the actual situation to improve the properties of wavelet transform.

和

Further, the forward transformation algorithm based on the improvement scheme of the optimization promotion process can be written as

and

Dual lift step

update step

After M pairs of update boosting steps and dual boosting steps, combined with the scale factors n _l and n _h , the even sample points become low-pass coefficients, and the odd sample points become high-pass coefficients, that is,

Usually, M is called the number of lifting steps, n _l and n _h are called normalization factors, and n ₁ ×n _h =1, for different biorthogonal wavelets, the values of n _l and n _h are synchronized.

Further, the parallel operation scheme of the encoder and the decoder of the SPIHT coding algorithm is completely foreseen from the DWT coefficient, that is, the importance of the pixel coefficient relative to the current threshold for the i-th bit plane is from the i+1th bit plane to the highest. The bit MSB is relatively independent between the bit planes in the logical OR operation of the coefficient bits of the pixel point, and the importance information of each coefficient in each bit plane can be obtained at the same time, realizing parallel processing of the planes.

Further, the encoder and decoder anti-interference fault tolerance scheme of the SPIHT encoding algorithm includes: a hyperspectral image encoding algorithm and a packed wavelet tree image compression algorithm;

The hyperspectral image coding algorithm divides the input data into multiple independent data packets with roughly the same length, and implements the coding algorithm in this paper for each data packet. The received data packets are used to approximately restore the lost information. The quality of image restoration depends on how many data packets are received, not on which specific packet is received;

The packaged wavelet tree image compression algorithm divides the code stream output by the wavelet zero-tree type encoder into fixed-length segments, and these segments can be decoded independently. Errors in one segment will not affect other segments, thereby realizing Anti-interference and fault tolerance.

Further, the basic steps of the SPIHT encoding algorithm are:

S1, initialization

置LSP为空，将坐标(i,j)∈H送入LIP，并将H中有后代(即高频部分：HLJ、LHJ、HHJ)的送入LIS，作为A型值output

Set LSP to be empty, send coordinates (i,j)∈H to LIP, and send those with descendants in H (ie high-frequency parts: HLJ, LHJ, HHJ) to LIS as the A-type value

S2, sorting process

S21. For each (i,j)∈LIP, output _Sn (i,j), if _Sn (i,j)=1, move (i,j) into LSP, and output C(i,j) )symbol;

S22. For each (i,j)∈LIS, do

S221. If it is an A-type value, then

①Output _Sn (D(i,j));

②If _Sn (D(i,j))=1, then for each (k,l)∈O(i,j), do:

output _Sn (k,l);

If _Sn (k,l)=1, send (k,l) into LSP and output its symbol;

If _Sn (k,l)=0, send (k,l) to the end of LIP;

③If L(i,j)≠φ, move (k,l) to the end of LIS as a B-type value; otherwise, delete (i,j) from LIS;

S222. If it is a B-type value, then

①Output _Sn (L(i,j));

②If _Sn (L(i,j))=1, then

Add to the end of LIS for each (k,l)∈O(i,j) as a type A value;

remove (i,j) from the LIS;

S3. Refinement process: output the nth most significant bit of |ci _,j | for each (i,j)∈LSP (excluding the most recent splitting process);

S4, quantization step refresh, n=n-1; return to S2;

The termination of encoding is determined by a given code rate. If it is lossless compression, it is encoded until n=0. When decoding, the output in the above algorithm only needs to be changed into the input.

The present invention has the following advantages:

The invention discloses a hyperspectral image lossless compression coding system, which adopts the three-band inter-spectral integer reversible transformation matrix of XCJRCT, combines the wavelet transform lifting scheme and the SPIHT encoder and decoder algorithm, and the simulation experiment verifies the Canal hyperspectral test image The time complexity of the SPIHT algorithm encoder and decoder algorithm, through the SPIHT compression coding experimental data of the first 160 bands of the hyperspectral standard test image canal.bsq, the SPIHT algorithm encoder and decoder algorithm of the Canal hyperspectral test image are realized. The time complexity test, by comparing the lossless compression comparison experiment of typical compression algorithms, verifies that the scheme in this paper is more efficient and has a higher compression ratio. The XCJRCT transform hyperspectral image lossless compression coding algorithm proposed in this paper, combined with the enhanced wavelet transform scheme, effectively realizes the lossless compression of hyperspectral images, and provides a high-efficiency algorithm for the hardware development of spectral detection sensors in the engineering field. Collect, transmit, identify and feedback environmental ecological factors.

Description of drawings

In order to illustrate the embodiments of the present invention or the technical solutions in the prior art more clearly, the following briefly introduces the accompanying drawings that are required to be used in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only exemplary, and for those of ordinary skill in the art, other implementation drawings can also be derived from the provided drawings without any creative effort.

The structures, proportions, sizes, etc. shown in this specification are only used to cooperate with the contents disclosed in the specification, so as to be understood and read by those who are familiar with the technology, and are not used to limit the conditions for the implementation of the present invention, so there is no technical The substantive meaning above, any modification of the structure, the change of the proportional relationship or the adjustment of the size should still fall within the technical content disclosed in the present invention without affecting the effect and the purpose that the present invention can produce. within the range that can be covered.

Fig. 1 is the realization process diagram of the normalization factor of the promotion scheme provided by the embodiment of the present invention;

FIG. 2 is a decomposition and reconstruction diagram of a lifting scheme provided by an embodiment of the present invention;

3 is a schematic diagram of a parallel encoder based on DWT and a bit plane provided by an embodiment of the present invention;

FIG. 4 is a schematic diagram of an anti-interference and fault tolerance solution for an encoder and a decoder provided by an embodiment of the present invention;

5 is a schematic diagram of a correlation coefficient of adjacent bands provided by an embodiment of the present invention;

FIG. 6 is a schematic diagram of a change curve of the lossless compression ratio of the first 160 bands of the hyperspectral image canal.bsq with the band according to an embodiment of the present invention.

Detailed ways

The embodiments of the present invention are described below by specific specific embodiments. Those who are familiar with the technology can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. Obviously, the described embodiments are part of the present invention. , not all examples. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

Example

This embodiment discloses a lossless compression coding system for hyperspectral images. The system utilizes parameterized three-band inter-spectral integer reversible transform combined with wavelet lifting scheme transform to remove inter-spectral transform, removes inter-spectral and spatial redundancy, and adopts SPIHT coding algorithm It realizes the lossless compression of hyperspectral images, and the de-redundant transformation includes: XCJRCT transform and wavelet transform promotion scheme. The SPIHT coding algorithm performs optimization analysis based on the time complexity algorithm through the parallel operation of the encoder and the decoder, and anti-interference and fault tolerance.

The adaptability and comprehensibility of the method are verified through numerical simulation and visualization studies in different fields. For any n×n invertible matrix A, if |det|=a≠1, regardless of whether the actual physical meaning corresponding to the transformation result is considered, that is, regardless of whether the modulus of the determinant value of the matrix is 1, the transformation matrix can always be transformed by transforming the matrix. Turn it into a modulo 1 matrix of determinant values. If the modulus of the determinant value of the matrix is equal to 1, then the matrix A has a finite basic triangular matrix decomposition, so a parameterized N-band inter-spectral integer reversible transformation matrix can be constructed. There are infinitely many such integer reversible transformations, which can be determined according to It is necessary to choose the optimal transform to actually deal with the problem. The parametric N-band inter-spectral integer reversible transform realized by the principle of the lowest energy is an example. It can be known from matrix theory that for a non-singular matrix A, there must be a triangular decomposition A=PLDU, where P, L, D, and U are the permutation matrix, unit lower triangular matrix, diagonal matrix and unit upper triangular matrix |detD|=| detPD|=|detPLDU|=|detA|=1, that is, the invertible transformation matrix A realized by n×n integers can be found. Although a structural proof is given for this research result, it is not parameterized, so it is impossible to talk about the optimal transformation problem. It can be proved that the matrix invertible transformations SHIRCT, RCT, YFbFr, YCbCr implemented by integers between the well-known three-band (R, G, B) spectra are not optimal. This is one of the main entry points of our research on this problem. Here, the optimal transformation solution problem is illustrated by taking the matrix reversible transformation realized by integers between three-band spectra as an example to show the importance of the research problem.

make

in,

Δ ₁ =a ₁ ×(1+c ₂ ×d ₁ )+a ₂ ×[b ₁ ×(1+c ₂ ×d ₁ )+b ₂ ×d ₁ ]+d ₁ , Δ ₂ =a ₁ ×( c ₁ +c ₂ ×d ₂ )+a ₂ ×[b ₁ ×(c ₁ +c ₂ ×d ₂ )+1+b ₂ ×d ₂ ]+d ₂ , Δ ₃ =a ₁ ×c ₂ +a ₂ ×(b ₁ ×c ₂ +b ₂ )+1, let T=PLAPR, where

but

It can be known from T in formula (1) and formula (2) that the first and second rows of the matrix have nothing to do with a ₁ and a ₂ , so the matrix reversible transformation realized by integer can be parameterized. b ₁ =-0.5, b ₂ =-0.1, c ₁ =1, c ₂ =0.5, d ₁ =-0.5, d ₂ =-0.5, we get

Δ ₁ =0.75×a ₁ -0.34375×a ₂ -0.5, Δ ₂ =0.96875×a ₁ +0.65625×a ₂ -0.5, Δ ₃ =0.5×a ₁ -0.3125×a ₂ +1. In order to make the transformation have the best effect, it is obvious that Δ ₁ +Δ ₂ +Δ ₃ =0, then a ₁ =0, in this way, the parameterized transformation matrix can be derived.

a ₂ =0.75, the famous SHIRCT transform is obtained.

Assuming Z=[Z ₁ , Z ₂ , Z ₃ ] ^T , X=[X ₁ , X ₂ , X ₃ ] ^T , then Z ₂ =(0.3125×X ₁ -0.65625×X ₂ +0.34375×X ₃ )× a ₂ +X ₁ -0.5×(X ₂ +X ₃ )=Z ₃ ×a ₂ +X ₁ -0.5×(X ₂ +X ₃ ), considering the actual effect of image compression, it is hoped that |Z ₂ | is small, because it is a parameterized expression, and Z ₂ and X ₁ -0.5×(X ₂ +X ₃ ) are fixed values, so the choice of a ₂ has a great influence on Z ₂ , and the difference of a ₂ Represents different reversible transformations. In theory, there are infinitely many such reversible transformations, but it is always hoped to find the optimal one. The optimal transformation can be selected according to the needs of the actual problem, such as the parametric three-band realized by the principle of minimum energy. Integer reversible transformation between spectra, etc. The explicit form XCJRCT transformation of the three-band inter-spectral integer reversible transformation matrix given in this embodiment is as follows:

Among them, γ and λ are parameters that can be adjusted during transformation, and their numerical values can be selected according to the actual needs of dealing with problems. In ₁ , In ₂ , and In ₃ are input signals, and Out ₁ , Out ₂ , and Out ₃ are output signals. '>>' is the binary left shift symbol. The above transformation can be completed by addition and shift, which is convenient for hardware implementation. The integer reversible transformation matrix between the 3-band spectra is written in the general parameter form as follows:

Choose an appropriate algorithm to get

Wherein, δ ₁ =x ₁ ×(1+z ₂ ×w ₁ )+x ₂ ×[y ₁ ×(z ₁ +z ₂ ×w ₂ )+y ₂ ×w ₂ +1]+w ₁ , δ ₂ =x ₁ ×(z ₁ +z ₂ ×w ₂ )+x ₂ ×[y ₁ ×(z ₁ +z ₂ ×w ₂ )+y ₂ ×w ₂ +1]+w ₂ , δ ₃ =x ₁ ×z ₂ +x ₂ ×(y ₁ ×z ₂ ×y ₂ )+1.

Therefore, γ and λ are adjustable parameters during transformation, and x ₁ , x ₂ , y ₁ , y ₂ , z ₁ , z ₂ , w ₁ and w ₂ are integer reversible transformation matrix formation parameters. The integer invertible transformation matrix of .

Wavelet transform DWT is currently recognized as a very good transform for removing spatial redundancy, and the knowledge obtained by DWT in removing spatial redundancy is quite mature, but using DWT to remove inter-spectral redundancy is not necessarily the best, even Very poor effect. For example, DWT is used to remove the inter-spectral redundancy of 7-band images, and only one decomposition is performed, which does not achieve the effect of de-redundancy, because the number of bands involved in DWT transformation is not a power of 2, which is not suitable for boundary continuation. Using DWT to remove the redundancy between spectra, because each extension point requires one more frame of image data, the amount of data is huge, which will have a serious adverse effect on the redundancy effect. If DWT must be used to remove inter-spectral redundancy, a hybrid transformation of DWT and other transformations can be used, DWT is used for the power of 2 part, and other transformations are used for the rest. Therefore, this paper proposes a unified parameterized integer reversible transform between N-band spectra. The parameterized integer reversible transform between N-band spectra is very flexible. It can achieve spectral de-redundancy alone, or can be combined with DWT or subtraction transform to achieve spectral to remove redundancy. In addition, another huge advantage is that the transformation can be completed by addition and shifting, which is fast and easy to implement in hardware.

S, TS and S+P transformations can all be regarded as special cases of the Swelden lifting scheme. The wavelet transform process implemented by the lifting scheme can be divided into four steps: splitting, predicting, updating, and optimizing the lifting step.

⑴ split

Split the original signal S _j,k into two mutually disjoint subsets S _j+1,k and d _j+1 , usually first perform Lazy wavelet or Polyphase wavelet transform on the original signal S _j,k , and transform a The signal sequence is divided into an even sequence and an odd sequence, ie split(S _j,k )=(S _j,2k ,S _j,2k+1 )=(S _j+1,k ,d _j+1,k ).

(2) Prediction

For the correlation between data, S _j+1,k can be used to predict d _j+1,k , so a prediction operator P that is independent of the data set structure can be used, so that d _j+1,k =P(S _{j +1,k} ), use the difference between d _j+1,k and the predicted value P(S _j+1,k ) to replace d _j+1,k , then this difference reflects the approximation degree of the two. If the predictions are reasonable, the difference dataset contains much less information than the original subset d _j+1,k .

⑶ update

Some overall properties (such as the mean) of the coefficient subset S _j+1,k generated by the above two steps are not consistent with the properties in the original data, so an update process is required. The purpose is to generate a better sub-data set S _j+1,k through the operator U to keep some characteristics of the original data set S _j,k , S _j+1,k is defined as S _{j+1, k} =S _j,2k+1 +U(d _j+1,k ). For the data subset S _j+1,k to perform the same splitting, prediction and update, S _j+1,k can be decomposed into d _j+2,k and S _j+2,k , after J times of decomposition, The wavelet transform of the original data S _0,k is expressed as {S _J ,d _J ,d _J-1 ,...,d ₁ }, where S _J represents the low-frequency part of the signal, and {d _J ,d _J-1 , ...,d ₁ } is the high frequency part of the signal.

⑷Optimize the lifting step

和

That is, according to the actual situation, the dual lifting step and the update lifting step are alternately applied to improve the properties of the wavelet transform. The forward transformation algorithm based on the lifting scheme can be written as

and

Dual lift step

update step

After M pairs of update boosting steps and dual boosting steps, combined with the scale factors n _l and n _h , the even sample points become low-pass coefficients, and the odd sample points become high-pass coefficients, that is,

Usually, M is called the number of lifting steps, n _l and n _h are called normalization factors, and n ₁ ×n _h =1, for different biorthogonal wavelets, the values of n _l and n _h are synchronized. Taking the three-level wavelet transform as an example, the normalization process after the orthogonal transform is shown in Figure 1.

The inverse transform can be written as

奇样点为

在实际应用中，通常利用提升方案实现整数-整数小波变换，即Finally, even samples and odd samples are obtained, that is, the even samples are

The odd point is

In practical applications, the integer-integer wavelet transform is usually implemented using a lifting scheme, i.e.

written as

为取整数运算，提升方案的分解和重构如图2所示。

For integer operations, the decomposition and reconstruction of the boosting scheme is shown in Figure 2.

In this embodiment, the parallel coding structure is studied, mainly to lay a solid foundation for the hardware implementation of the encoder. The study finds that the bit planes transformed by DWT are correlated with each other through amplitudes, and can be completely predicted from the DWT coefficients, that is, for The importance of the pixel coefficient of the i-th bit plane relative to the current threshold is the logical OR operation of the i+1-th bit plane to the most significant MSB in the pixel coefficient bits. Therefore, it can be considered that the bit planes are relatively independent, and the importance information of each coefficient in each bit plane can be obtained at the same time, so that the parallel processing of the bit planes becomes possible. The parallel encoder based on DWT and bit planes is shown in Figure 3. Show. The coding algorithm in Fig. 3 adopts SPIHT, and the SPIHT coding algorithm is appropriately modified according to the parallel coding tree structure in the specific implementation.

By studying the anti-interference and fault tolerance of compression coding, a highly robust fault-tolerant coding algorithm can be realized. The specific scheme is realized in the following two aspects.

⑴Hyperspectral image coding algorithm

Divide the input data into multiple independent data packets with approximately the same length, and implement the encoding algorithm in this paper for each data packet. When some of the packets are lost, the receiving end relies on the received data packets to approximately recover the lost information. The quality of image restoration depends on how many packets are received, not on which packets are received.

(2) Packed wavelet zero-tree image compression algorithm

The code stream output by the wavelet zero-tree encoder is divided into fixed-length segments. These segments can be decoded independently. Errors in one segment will not affect other segments, so as to achieve anti-interference and fault tolerance functions. Combined with this paper, the anti-jamming and fault-tolerant scheme of coding algorithm is proposed. The anti-interference and fault-tolerant scheme of the encoder and decoder is shown in Figure 4.

The basic steps of the SPIHT encoding algorithm are:

S1, initialization

置LSP为空，将坐标(i,j)∈H送入LIP，并将H中有后代(即高频部分：HLJ、LHJ、HHJ)的送入LIS，作为A型值output

Set LSP to be empty, send coordinates (i,j)∈H to LIP, and send those with descendants in H (ie high-frequency parts: HLJ, LHJ, HHJ) to LIS as the A-type value

S2, sorting process

S21. For each (i,j)∈LIP, output _Sn (i,j), if _Sn (i,j)=1, move (i,j) into LSP, and output C(i,j) )symbol;

S22. For each (i,j)∈LIS, do

S221. If it is an A-type value, then

①Output _Sn (D(i,j));

②If _Sn (D( _i,j ))=1, then for each (k,l)∈O(i,j), do:

output _Sn (k,l);

If _Sn (k,l)=1, send (k,l) into LSP and output its symbol;

If _Sn (k,l)=0, send (k,l) to the end of LIP;

③If L(i,j)≠φ, move (k,l) to the end of LIS as a B-type value; otherwise, delete (i,j) from LIS;

S222. If it is a B-type value, then

①Output _Sn (L(i,j));

②If _Sn (L(i,j))=1, then

Add to the end of LIS for each (k,l)∈O(i,j) as a type A value;

remove (i,j) from the LIS;

S3. Refinement process: output the nth most significant bit of |ci _,j | for each (i,j)∈LSP (excluding the most recent splitting process);

S4, quantization step refresh, n=n-1; return to S2;

The termination of encoding is determined by a given code rate. If it is lossless compression, it is encoded until n=0. When decoding, the output in the above algorithm only needs to be changed into the input.

In this embodiment, in order to study the storage complexity and time complexity of the encoder and decoder algorithms, under the premise of optimizing the memory storage, all the encoders and decoders are tested under the same test conditions. Algorithms were tested. This paper uses Canal hyperspectral images for testing, and provides 24 3-band color standard test images (768×512×24bits) for testing. The time complexity test of the SPIHT algorithm encoder and decoder algorithm is shown in Table 1.

Table 1 Time complexity test of SPIHT algorithm encoder and decoder algorithm for Canal hyperspectral test image

The test results in Table 1 verify that the time complexity of the Canal hyperspectral test image SPIHT algorithm encoder and decoder algorithm are very different.

In this embodiment, the hyperspectral image canal.bsq tested by the simulation experiment has a spectral resolution of 10 nm, a spectral range of 400 nm-2400 nm, a total of 223 bands, the spectral cross-correlation coefficient Ri of adjacent bands of canal.bsq, and the correlation between adjacent bands. The coefficients are shown in Figure 5.

In Figure 5, the spectral correlations between bands 106-113, bands 152-157, and bands 217-222 have a very obvious decrease. The entire hyperspectral image is divided into 5 parts, which can be transformed according to different groupings of spectral band correlations. Through analysis, it is found that the published predictions and transformations to remove spectral redundancy have not effectively improved the compression ratio. Therefore, it is necessary to study segment-based matrix transformation to remove spectral redundancy and improve the effect of hyperspectral image compression.

Therefore, this paper adopts the 160-band canal.bsq hyperspectral remote sensing image for compression simulation experiment. The hyperspectral image is de-redundant by using inter-spectral and spatial joint transformation to achieve de-redundancy. Inter-spectral transformation adopts XCJRCT transformation, and intra-frame transformation adopts CDF (2, 2) lifting scheme wavelet transformation, and then implements SPIHT coding on the transformation result. , Table 2 shows the experimental results of SPIHT compression coding for the first 160 bands of the hyperspectral remote sensing test image canal.bsq.

Table 2 Experimental results of SPIHT compression coding for the first 160 bands of hyperspectral image canal.bsq

This embodiment is compared with other typical lossless compression algorithms. JPEG-LS is a lossless compression algorithm based on the LOCO-I algorithm. It effectively reduces the entropy of the error image through the context model and error feedback, and then realizes the error image through run-length coding. 's encoding. The 160 bands are divided into 10 groups, 1D-CDF(2,2) DWT is used between the spectra, and 2D-CDF(2,2) DWT is used for SPIHT compression coding in the frame. The lossless compression ratio of the first 160 bands of the hyperspectral image canal.bsq varies with The band change curve is shown in Figure 6.

The average lossless compression ratio of the 160-band is 2.728695; the lossless compression ratio of the 160-band varies with the band. WinZip is a well-known lossless image compression method proposed by Microsoft. The ARJ method adopts a single-pass adaptive Huffman lossless compression algorithm. DPCM is a typical prediction-based lossless compression algorithm. The experimental results of the comparison between the algorithm in this paper and the typical compression algorithm are shown in Table 3.

Table 3 The experimental results of the comparison between the algorithm in this paper and the typical compression algorithm

It can be seen from the experimental results that the lossless compression ratio of each band changes faster with the band, the compression ratio changes in the range of [1.5984, 4.1686], and the compression ratio is 2.728695. The transformation scheme is higher than typical algorithms such as JPEG-LS, WinZip, ARJ and DPCM, respectively. 73.692%, 67.713%, 65.175%, 59.108%. Therefore, adopting the XCJRCT wavelet lifting scheme is very effective.

Although the present invention has been described in detail above with general description and specific embodiments, some modifications or improvements can be made on the basis of the present invention, which will be obvious to those skilled in the art. Therefore, these modifications or improvements made without departing from the spirit of the present invention fall within the scope of the claimed protection of the present invention.

Claims (7)

1. a hyperspectral image lossless compression coding system, is characterized in that, described system utilizes parameterized three-band inter-spectral integer reversible transform in conjunction with wavelet lifting scheme transformation to remove inter-spectral transform, removes between spectrum and spatial redundancy, adopts SPIHT encoding The algorithm realizes the lossless compression of hyperspectral images, and the de-redundant transformation includes: XCJRCT transform and wavelet transform promotion scheme. The SPIHT coding algorithm performs optimization analysis based on the time complexity algorithm through the parallel operation of the encoder and the decoder, and anti-interference and fault tolerance. 2. a kind of hyperspectral image lossless compression coding system as claimed in claim 1, is characterized in that, described XCJRCT transform realizes matrix reversible transformation with the integer between three-band spectrum, carries out the optimal transformation solution, and the integer between three-band spectrum is reversible. The transformation matrix display form XCJRCT transforms as:

Among them, γ and λ are parameters that can be adjusted during transformation, and their numerical values can be selected according to the actual needs of dealing with problems. In ₁ , In ₂ , and In ₃ are input signals, and Out ₁ , Out ₂ , and Out ₃ are output signals. '>>' is the binary left shift symbol. 3. a kind of hyperspectral image lossless compression coding system as claimed in claim 1 is characterized in that, described wavelet transform promotion scheme comprises: splitting, predicting, updating and optimizing promotion; Splitting, splitting the original signal S _j,k into two mutually disjoint subsets S _j+1,k and d _j+1 , usually first perform Lazy wavelet or Polyphase wavelet transform on the original signal S _j,k , and transform An original signal sequence is divided into an even sequence and an odd sequence, namely split(S _j,k )=(S _j,2k ,S _j,2k+1 )=(S _j+1,k ,d _j+1,k ); Prediction, for the correlation between data, S _j+1,k can be used to predict d _j+1,k , so a prediction operator P that is independent of the data set structure can be used, so that d _j+1,k =P( S _j+1,k ), use the difference between d _j+1,k and the predicted value P(S _j+1,k ) to replace d _j+1,k , then this difference reflects the approximation degree of the two , if the prediction is reasonable, the difference data set contains less information than the original subset d _j+1,k ; Update, some overall properties of the coefficient subset S _j+1,k generated by the two steps of splitting and prediction are not consistent with the properties of the original data, and an update process is required to generate a better sub-set through the operator U. The data set S _j+1,k keeps some characteristics of the original data set S _{j,k ,} the definition of S _{j+1,k is S j+1,k} =S _j,2k+1 +U(d _{j +1,k} ), perform the same splitting, prediction and update for the data subset S _j+1,k, then S _j+1,k can be decomposed into d _j+2,k and S _j+2,k , After decomposing J times, the wavelet transform of the original data S _0,k is expressed as {S _J ,d _J ,d _J-1 ,...,d ₁ }, where S _J represents the low-frequency part of the signal, and {d _J ,d _J-1 ,...,d ₁ } is the high frequency part of the signal, Optimized lifting, applying dual lifting steps and update lifting steps alternately according to the actual situation to improve the properties of wavelet transform. 4. a kind of hyperspectral image lossless compression coding system as claimed in claim 3 is characterized in that, the forward transformation algorithm of described optimization promotion process based on promotion scheme can be written as

and

Dual lift step

update step

After M pairs of update boosting steps and dual boosting steps, combined with the scale factors n _l and n _h , the even sample points become low-pass coefficients, and the odd sample points become high-pass coefficients, that is,

Usually, M is called the number of lifting steps, n _l and n _h are called normalization factors, and n ₁ ×n _h =1, for different biorthogonal wavelets, the values of n _l and n _h are synchronized. 5. a kind of hyperspectral image lossless compression coding system as claimed in claim 1, is characterized in that, the encoder of described SPIHT coding algorithm, the parallel operation scheme of decoder completely foresee from DWT coefficient, namely for the ith bit plane The importance of the pixel coefficient relative to the current threshold is the logical OR operation of the i+1th bit plane to the highest bit MSB in the pixel coefficient bits. The bit planes are relatively independent, and each coefficient is in each bit plane. The importance information can be obtained at the same time, realizing the parallel processing of the plane. 6. a kind of hyperspectral image lossless compression coding system as claimed in claim 1 is characterized in that, the encoder of described SPIHT coding algorithm, decoder anti-interference fault tolerance scheme comprise: hyperspectral image coding algorithm and packing wavelet tree image compression algorithm; The hyperspectral image coding algorithm divides the input data into multiple independent data packets with roughly the same length, and implements the coding algorithm in this paper for each data packet. The received data packets are used to approximately restore the lost information. The quality of image restoration depends on how many data packets are received, not on which specific packet is received; The packaged wavelet tree image compression algorithm divides the code stream output by the wavelet zero-tree type encoder into fixed-length segments, and these segments can be decoded independently. Errors in one segment will not affect other segments, thereby realizing Anti-interference and fault tolerance. 7. a kind of hyperspectral image lossless compression coding system as claimed in claim 1, is characterized in that, described SPIHT coding algorithm basic step is: S1, initialization output

Set LSP to be empty, send coordinates (i,j)∈H to LIP, and send those with descendants in H (ie high-frequency parts: HLJ, LHJ, HHJ) to LIS as the A-type value S2, sorting process S21. For each (i,j)∈LIP, output _Sn (i,j), if _Sn (i,j)=1, move (i,j) into LSP, and output C(i,j) )symbol; S22. For each (i,j)∈LIS, do S221. If it is an A-type value, then ①Output _Sn (D(i,j)); ②If _Sn (D(i,j))=1, then for each (k,l)∈O(i,j), do: output _Sn (k,l); If Sn(k,l)=1, send (k,l) into LSP and output its symbol; If Sn(k,l)=0, send (k,l) to the end of LIP; ③If L(i,j)≠φ, move (k,l) to the end of LIS as a B-type value; otherwise, delete (i,j) from LIS; S222. If it is a B-type value, then ①Output _Sn (L(i,j)); ②If _Sn (L(i,j))=1, then Add to the end of LIS for each (k,l)∈O(i,j) as a type A value; remove (i,j) from the LIS; S3. Refinement process: output the nth most significant bit of |ci _,j | for each (i,j)∈LSP (excluding the most recent splitting process); S4, quantization step refresh, n=n-1; return to S2; The termination of encoding is determined by a given code rate. If it is lossless compression, it is encoded until n=0. When decoding, the output in the above algorithm only needs to be changed into the input.

Images