CN104639947B - High spectrum image compression method based on classification discrete cosine transform - Google Patents
High spectrum image compression method based on classification discrete cosine transform Download PDFInfo
- Publication number
- CN104639947B CN104639947B CN201510097077.XA CN201510097077A CN104639947B CN 104639947 B CN104639947 B CN 104639947B CN 201510097077 A CN201510097077 A CN 201510097077A CN 104639947 B CN104639947 B CN 104639947B
- Authority
- CN
- China
- Prior art keywords
- spectral
- vector
- spectrum
- class
- 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.)
- Active
Links
- 238000001228 spectrum Methods 0.000 title claims abstract description 64
- 238000007906 compression Methods 0.000 title claims abstract description 41
- 230000006835 compression Effects 0.000 title claims abstract description 41
- 238000000034 method Methods 0.000 title claims abstract description 31
- 239000013598 vector Substances 0.000 claims abstract description 118
- 230000009466 transformation Effects 0.000 claims abstract description 25
- 230000003595 spectral effect Effects 0.000 claims description 96
- 230000001131 transforming effect Effects 0.000 claims 1
- 238000012545 processing Methods 0.000 abstract description 5
- 230000005540 biological transmission Effects 0.000 abstract 1
- 238000004088 simulation Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 5
- 238000003384 imaging method Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000011426 transformation method Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- BERDEBHAJNAUOM-UHFFFAOYSA-N copper(I) oxide Inorganic materials [Cu]O[Cu] BERDEBHAJNAUOM-UHFFFAOYSA-N 0.000 description 1
- LBJNMUFDOHXDFG-UHFFFAOYSA-N copper;hydrate Chemical compound O.[Cu].[Cu] LBJNMUFDOHXDFG-UHFFFAOYSA-N 0.000 description 1
- 238000013144 data compression Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008707 rearrangement Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Landscapes
- Compression, Expansion, Code Conversion, And Decoders (AREA)
Abstract
The invention discloses a kind of high spectrum image compression method based on classification discrete cosine transform, mainly solves prior art and does not consider spectrum vector correlation when convert between spectrum, causes compression decorrelation not thorough, the problem of compression effectiveness difference.Implementation step is：1) space two-dimensional wavelet transformation is carried out to high spectrum image；2) the spectrum vector that spatial wavelet transform coefficient is formed is classified using the sorting algorithm based on spectrum vectorial property, obtains classification chart, and such mean value vector is subtracted to every a kind of spectrum vector according to classification chart, obtain residual error vector；3) line translation is entered to residual error vector using one-dimensional dct transform between spectrum, obtains three-dimension varying coefficient；4) three-dimension varying coefficient is encoded, obtains the compressed bit stream of code check controllable precise.The present invention makes full use of the statistical property between high spectrum image spectrum, makes decorrelation more thorough, and more preferable compression performance is obtained under equal code check, available for hyperspectral data processing and transmission.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to a hyperspectral image lossy compression method which can be used for processing and transmitting various hyperspectral data.
Background
The hyperspectral image is a three-dimensional data cube which is obtained by imaging the same ground object on hundreds of spectral bands by an imaging spectrometer and simultaneously contains spatial information and spectral information, and is widely applied to the aspects of resource exploration, target identification, environmental protection and the like. Since the amount of hyperspectral image data is huge, an effective compression technology needs to be adopted for storing and transmitting the image. Especially, in a satellite-borne hyperspectral image compression system, due to the limitation of satellite channel bandwidth, it is difficult to transmit such a large data volume in real time, and therefore lossy compression is often performed on hyperspectral images.
Among the existing lossy compression methods, transform-based compression methods such as the A3D iphiht algorithm (asymmetric Three-dimensional multi-level tree set splitting algorithm, "image processing,2003, ICIP 2003, proceedings.2003international conference. vol.2.ieee, 2003), 3D peck (Three-dimensional set splitting insert, Tang, Xiaoli, and williama. pearl." Three-dimensional spatial way-based compression of Hyperspectral Data compression, spring US,2006.273-308, etc. are the most typical. The A3DSPIHT algorithm firstly carries out asymmetrical 3-dimensional wavelet transform DWT on an image, namely, spatial two-dimensional wavelet transform is combined with spectral one-dimensional wavelet transform, and then codes the obtained transform coefficient by using the 3DSPIHT algorithm, so that a better compression effect can be obtained. The 3DSPECK algorithm is similar to A3DSPIHT, and asymmetric 3-dimensional wavelet transform is performed on the image at first, the difference is that the 3DSPECK algorithm is adopted for encoding the transform coefficient, and the compression efficiency is slightly improved compared with the A3 DSPIHT. However, these methods all use one-dimensional wavelet transform between spectra, and the decorrelation capability between spectra is limited, so some researchers have proposed to replace wavelet transform with KLT transform to better remove the correlation between spectra and greatly improve the compression performance, such as KLT +3DSPECK, which uses two-dimensional wavelet transform in the spatial domain, and one-dimensional KLT transform between spectra, and finally uses 3DSPECK algorithm for compression. However, when the KLT is used for inter-spectral transformation, the statistical characteristics of the image are required to be known, the algorithm complexity is high, and a uniform inter-spectral transformation matrix is usually used, so that the optimal decorrelation performance cannot be obtained, and if the inter-spectral decorrelation performance is improved by using a method of blocking or classifying the image, the algorithm complexity is further increased.
In theory, DCT transform is the closest transform method to KLT transform and the complexity is much lower than that of KLT transform, so in order to take account of both algorithm complexity and coding performance, some researchers have studied how to achieve performance closer to that of KLT transform using DCT transform, and proposed compression algorithm based on categorical Residual DCT, see Zhang, king, and Guizhong liu, "a noveltaibed Residual DCT For Hyperspectral Images scale compression. In the hyperspectral images, a spectral vector refers to a one-dimensional vector formed by values of pixels in the same spatial position and different spectral bands in a group of hyperspectral images. The compression algorithm based on the classification residual DCT adopts two-dimensional wavelet transform during space transform, and during inter-spectrum transform, firstly classifies all spectrum vectors according to the size of the mean value of the spectrum vectors, then subtracts the mean value vector of the class from the spectrum vectors in each class, then carries out one-dimensional DCT transform on the obtained residual vectors, and finally carries out 3DSPIHT coding on transform coefficients. The characteristic of a spectral vector is effectively combined in the process of inter-spectral transformation of the classification residual DCT transformation method, the transformation performance of the classification residual DCT transformation method is superior to that of DCT and DWT and is closer to KLT, so that the performance of a compression algorithm based on the classification residual DCT is superior to that of A3DSPIHT and is close to that of KLT +3DSPECK, but the complexity is lower. However, when the method classifies spectral vectors, only a simple spectral vector mean is used as a classification basis, and the mean cannot completely describe the characteristics of the spectral vectors, so that the classification is inaccurate, the classification precision is low, and the transformation performance and the final compression effect are affected.
Disclosure of Invention
The invention aims to provide a hyperspectral image lossy compression method based on classified discrete cosine transform aiming at the problem that the classification of spectral vectors is inaccurate when a compression algorithm based on classified residual DCT is subjected to inter-spectral transform, so as to improve the classification precision of the spectral vectors and obtain a better coding effect under the condition of lower coding complexity.
In order to achieve the purpose, the technical scheme of the invention comprises the following steps:
(1) inputting a hyperspectral image (W) with the total spectral band number of Y1,W2,…,Wk,…,WYEqually dividing the hyperspectral image into N groups according to the sequence of the spectral segments to obtain grouped hyperspectral images, and using the hyperspectral imagesRepresenting an m-th set of hyperspectral images, wherein,representing the k-th spectral band image in the m-th group,the real pixel gray values of the kth spectral band, the ith row and the jth column in the mth group of hyperspectral images are represented, i is 1,2, …, H, j is 1,2, …, W, k is 1,2, …, P, W is the image width, H is the image height, P represents the number of spectral bands of the input group of hyperspectral images, and P is Y/N;
(2) for the m group of the grouped hyperspectral imagesPerforming spatial wavelet transform on each spectral band to obtain the mth group of spatial wavelet transform coefficientsWhereinRepresenting a transform coefficient after wavelet transform of a kth spectral band image in the mth group of hyperspectral images;
(3) for the mth group of space wavelet transformation coefficientsAnd (4) classifying:
(3a) inputting the spatial wavelet transform coefficient obtained by the m-th group of hyperspectral image transformAccording to the spectral vectorThe size of the mean is classified to obtain an initial classification map p1 with height H and width W, whereinRepresenting the values of the spatial wavelet transform coefficients of the mth group of hyperspectral images on the kth spectral band of the ith row and the jth column;
(3b) by utilizing the variation range of each spectrum vector of each class in the initial classification map p1, dividing the spectrum vectors of which the variation ranges are in the same interval into one class to obtain a refined classification map p2 with the height of H and the width of W;
(3c) carrying out final classification by using the distribution positions of the maximum value and the minimum value of each spectrum vector in the refined classification map p2, classifying the spectrum vectors with the same positions of the maximum value and the minimum value into one class to obtain a final classification map p3 with the height of H and the width of W, and simultaneously recording the number L of the classes of the final classification;
(4) calculating the mean vector of the classified spectral vectors according to the classes by using the final classification map p3Subtracting the mean vector of the class from the spectrum vector of each class to obtain the residual vector of the spectrum vectorWhereinRepresents the mean vector of all spectral vectors in class i, L-0, 1,2, …, L,the index i, j of (a) indicates the spectral vector corresponding to the residual vectorMeasuring the rows and columns in the image, indicating the category of the residual vector by the superscript l, finally outputting the residual vector to the inter-spectrum transformation unit, and finally classifying the final classification map p3 and the mean vector corresponding to each categoryThe side information is stored and transmitted to a decoding end;
(5) for residual error vectorAnd performing one-dimensional DCT (discrete cosine transformation) to obtain a DCT coefficient, outputting the DCT coefficient serving as a final transformation result to a coding unit of a three-dimensional multi-level tree set splitting algorithm 3DSPIHT to obtain a compressed code stream file with an accurately controllable code rate, and ending coding.
The invention carries out classification pretreatment before the inter-spectrum transformation, adopts a classification mode of three steps of gradual thinning, fully utilizes the characteristics of the spectrum vectors, improves the classification accuracy and ensures stronger correlation among the spectrum vectors in the same classification after classification compared with a classification method of simply adopting a mean value, thereby obtaining the spectrum vector residual error with smaller amplitude and effectively improving the compression performance.
Drawings
FIG. 1 is a general flow chart of an implementation of the present invention;
fig. 2 is a sub-flow diagram of the inter-spectral classification in the present invention.
Detailed Description
The present invention is described in further detail below with reference to the attached drawings.
Referring to fig. 1, the implementation steps of the invention are as follows:
step 1, inputting a hyperspectral image.
Inputting a hyperspectral image { W with total spectral segments of Y, width of W and height of H1,W2,…,Wy,…,WYIn which W isy＝{I1,1,y,I1,2,y,…,Ii,j,y,…,IH,W,yDenotes the y-th spectral band image, Ii,j,yReal pixel gray values of the ith spectrum band, the ith row and the jth column in the hyperspectral image are represented, i is 1,2, …, H, j is 1,2, …, W, Y is 1,2, … and Y;
and 2, grouping the input hyperspectral images.
To be inputted hyperspectral image { W1,W2,…,Wy,…,WYEqually dividing the image into N groups to obtain grouped hyperspectral images, and using the hyperspectral imagesRepresenting a mth set of hyperspectral images, whereinRepresenting the k-th spectral band image in the m-th group,the real pixel gray values of the kth spectral band, the ith row and the jth column in the mth group of hyperspectral images are represented, i is 1,2, …, H, j is 1,2, …, W, k is 1,2, …, P and P represent the number of spectral bands of the input group of hyperspectral images, and P is Y/N;
the grouping of the hyperspectral images can be realized by adopting various existing methods, such as grouping according to the correlation among the spectrums, spectrum segment rearrangement, and the like, but the grouping is not limited to be performed according to the sequence of the spectrum segments, and the grouping is not limited to be P-16.
And 3, performing spatial wavelet transformation on the grouped hyperspectral images.
For the m group of the grouped hyperspectral imagesPerforming spatial wavelet transform on each spectral band to obtain the mth group of spatial wavelet transform coefficientsAnd output thereinRepresenting a transform coefficient after wavelet transform of a kth spectral band image in the mth group of hyperspectral images;
the pair of input hyperspectral imagesThe wavelet transform is generally performed by directly using 9/7 wavelets, and in this embodiment, before performing the 9/7 wavelet transform, all pixel points are subjected to mean value removal processing to reduce the size of pixel values and improve the precision of the wavelet transform.
Step 4, for the input mth group space wavelet transformation coefficientAnd carrying out spectrum classification to obtain a classification graph and a classification residual error.
Said pair of m-th group space wavelet transform coefficientsThe inter-spectrum classification can be realized by various existing methods, such as mean-based classification, wavelet sub-band-based classification, etc., the present embodiment proposes and adopts but is not limited to a spectral-inter-spectrum classification method based on three steps of mean, variation range and variation trend of the spectral vector,
referring to fig. 2, this step is implemented as follows:
(4a) inputting the spatial wavelet transform coefficient obtained by the m-th group of hyperspectral image transform, and according to the spectral vectorClassifying the mean value to obtain an initial classification map p1 with the height H and the width W;
(4a1) taking different spectral bands of space transformation coefficient, the values at the same row and column form a spectral vector WhereinRepresenting the values of the spatial wavelet transform coefficients of the mth group of hyperspectral images on the kth spectral band of the ith row and the jth column;
(4a2) finding each spectral vectorMean value ofWherein,representing the mean value of the spectral vectors of the ith row and the jth column in the mth group of spatial wavelet transform coefficients;
(4a3) for the mean value of each spectral vectorAnd (4) classifying:
first, the maximum variation range of the absolute values of all spectral vectors in the m-th group is determinedAnd dividing the range into a set Th of intervals with an integer power of 2 as a threshold:
then, aiming at the characteristic that the variation range of the interval at 0 is very small, a mean value combining threshold Tm is selected, and the range is [ -2 ]Tm,2Tm]The classes are combined into one class, and a set Th' of combined intervals is obtained: wherein Tm is determined from the gray value of the compressed image;
finally, according to the interval divided by the combined interval set Th', the average value of each spectrum vector in the same intervalAre classified into one category;
(4b) by utilizing the variation range of each spectrum vector of each class in the initial classification map p1, dividing the spectrum vectors of which the variation ranges are in the same interval into one class to obtain a refined classification map p2 with the height of H and the width of W;
(4b1) calculating each spectral vectorThe variation range of (2):whereinRepresenting the variation range of the spectral vector at the ith row and the jth column in the mth group, wherein k is the number of spectral segments;
(4b2) according to the initial classification map p1, fromIn finding out the person who belongs toAll spectral vector ranges of the same class, usingWherein L is 0,1,2, …, L1，L1Is the total number of categories in the initial classification map p 1;
(4b3) according to the variation range of each spectral vector in class lAll spectral vectors in this category are classified:
first, all the parameters are obtainedMaximum range of variation of absolute valueAnd dividing the range into a plurality of intervals by taking the integral power of 2 as a threshold:
Th1is a collection of partitioned areas;
then, aiming at the characteristic that the variation range of the interval at 0 is very small, a variation range combination threshold Td is selected, and the range combination threshold Td is [ -2 ]Td,2Td]The categories in between are combined into one category:Th1' is a set of merged intervals;
finally, according to Th1' divided interval, spectral vector is varied in rangeSpectral vectors in the same interval are classified into one type;
(4c) the distribution positions of the maximum value and the minimum value of each spectrum vector of each class in the refined classification map p2 are used for final classification:
(4c1) finding spectral vectors from the refined classification map p2And finding the positions of the maximum component and the minimum component of each spectral vector in the class in the spectral bands:andwherein k is1The spectral sequence number, k, representing the maximum component of the spectral vector2The sequence number, k, of the spectral band in which the minimum component of the spectral vector is located1And k2The following four cases are distributed:
(4c2) for all the spectral vectors in the same class obtained in the previous step, the sequence number k of the spectral segment where the maximum value component is located1And the spectral sequence number k of the minimum component2The spectral vectors belonging to the same situation are classified into one class, a final classification map p3 with the height H and the width W is obtained, and the number L of the finally classified classes is recorded.
And 5, solving a residual vector of the classified spectral vectors.
(5a) Finding out corresponding spectrum vectors according to categories by utilizing the final classification map p3, and then calculating the mean vector of the spectrum vectors in each categoryWhereinMean vector representing all spectral vectors in class i, L ═ 0,1,2, …, L;
(5b) subtract the mean vector of the class in which it is located from the spectral vector in class IObtaining the residual error vector of the spectrum vectorWhereinThe subscript i, j of (a) indicates the row and column of the spectrum vector corresponding to the residual vector in the image, and the subscript l indicates the category of the residual vector;
(5c) the final classification map p3 and the mean vectorStored in the form of side information and transmitted to the decoding end.
Step 6, residual error vector is processedAnd performing one-dimensional DCT to obtain a DCT coefficient F (u):
where f (x) represents the input signal source, and the coefficient C (u) is defined as:
step 7, recombining the one-dimensional DCT transform coefficients F (u) according to the spectral bands, and arranging the one-dimensional DCT transform coefficients F (u) according to the corresponding positions of the corresponding spectral vectors in the original hyperspectral image to obtain a three-dimensional final transform coefficient
Step 8, the final transformation coefficient is processedAnd coding to obtain a compressed code stream file with accurately controllable code rate, and finishing coding.
The pair of final transform coefficientsThe encoding can be implemented by using various existing methods, such as an embedded encoding method EZW, a three-dimensional set splitting embedded block 3DSPECK, and the like, and the final transform coefficient is encoded by using, but not limited to, a three-dimensional multi-level tree set splitting algorithm 3 dspith in the embodiment.
The effects of the present invention can be further explained by the following simulation experiments.
(1) Simulation conditions
The software is implemented on the Windows7 environment of Microsoft corporation using Microsoft Visual Studio 2010 integration development software and C language. The invention selects an AVIRIS hyperspectral standard image. The data are acquired on the wavelength of 0.4-2.5 um, the total number of the data is 224 spectrum sections, the spectral resolution is 10nm, the spatial resolution is 20m × 20m, each pixel is stored in 2 bytes, the selected 4 scenes are 'Jasper Ridge', 'Cuprite', 'Lunar Lake' and 'LowAltitude', the bit depth is 16 bits, and the size of the spectrum section is 512 pixels.
(2) Emulated content
The invention and the existing KLT +3DSPECK method, the existing classification residual DCT method, the existing DCT +3DSPIHT method and the existing A3DSPIHT method are respectively utilized to carry out compression simulation experiments on the four AVIRIS hyperspectral standard images, the SNR simulation results of the four AVIRIS hyperspectral standard images are shown in the table 1, and the complexity simulation results are shown in the table 2.
In tables 1 and 2, each behavior is a result of a different compression method, each behavior indicates a result under a different code rate, the code rate is bpppb, and indicates a bit rate of each pixel point of each spectrum, for example, 0.5 indicates that each pixel point of each spectrum occupies 0.5 bits in a compressed code stream after compression. Each entry in table 2 is the average encoding time of 4 hyperspectral images at different code rates, and the unit is ms.
TABLE 1 AVIRIS image simulation SNR (Unit: dB)
As can be seen from table 1, the present invention can obtain good compression effects at both high compression ratio and low compression ratio. The work of removing the inter-spectrum correlation of the hyperspectral image is mainly concentrated on the inter-spectrum classification unit, and the classification result is transmitted to the decoding end in the form of side information, and the method is different from the traditional method that the inter-spectrum correlation is completely concentrated on the inter-spectrum transformation module, so that the decorrelation is more thorough, and the result is not easily influenced by the compression ratio of the final coding coefficient, and the compression performance can be further improved.
TABLE 2 complexity comparison (in ms) of several compression schemes
As can be seen from table 2, the present invention has stable computational complexity at each compression ratio. Because the invention transfers the complex spatial decorrelation process from the traditional inter-spectrum transformation to the joint completion of the inter-spectrum classification and the relatively simple one-dimensional DCT transformation, the complexity of the inter-spectrum transformation is greatly simplified, the coding complexity is not obviously increased on the high compression ratio, and the complexity of each compression ratio is more uniform and stable.
The above description is only one specific example of the present invention and should not be construed as limiting the invention in any way. It will be apparent to persons skilled in the relevant art(s) that, having the benefit of this disclosure and the teachings and principles disclosed herein, numerous modifications and variations in form and detail can be made without departing from the principles and structures of the invention, which are set forth in the following claims.
Claims (4)
1. A hyperspectral image lossy compression method based on classified discrete cosine transform comprises the following steps:
(1) inputting a hyperspectral image (W) with the total spectral band number of Y1,W2,…,Wy,…,WYEqually dividing the hyperspectral image into N groups to obtain grouped hyperspectral imagesRepresenting an m-th set of hyperspectral images, wherein,representing the k-th spectral band image in the m-th group,the real pixel gray values of the kth spectral band, the ith row and the jth column in the mth group of hyperspectral images are represented, i is 1,2, …, H, j is 1,2, …, W, k is 1,2, …, P, W is the image width, H is the image height, P represents the number of spectral bands of the input group of hyperspectral images, and P is Y/N;
(2) for the m group of the grouped hyperspectral imagesPerforming spatial wavelet transform on each spectral band to obtain the mth group of spatial wavelet transform coefficientsWhereinRepresenting a transform coefficient after wavelet transform of a kth spectral band image in the mth group of hyperspectral images;
(3) for the mth group of space wavelet transformation coefficientsAnd (4) classifying:
(3a) inputting the spatial wavelet transform coefficient obtained by the m-th group of hyperspectral image transform, and according to the spectral vectorThe size of the mean is classified to obtain an initial classification map p1 with height H and width W, whereinRepresenting the values of the spatial wavelet transform coefficients of the mth group of hyperspectral images on the kth spectral band of the ith row and the jth column;
(3b) by utilizing the variation range of each spectrum vector of each class in the initial classification map p1, dividing the spectrum vectors of which the variation ranges are in the same interval into one class to obtain a refined classification map p2 with the height of H and the width of W;
(3c) carrying out final classification by using the distribution positions of the maximum value and the minimum value of each spectrum vector in the refined classification map p2, classifying the spectrum vectors with the same positions of the maximum value and the minimum value into one class to obtain a final classification map p3 with the height of H and the width of W, and simultaneously recording the number L of the classes of the final classification;
(4) calculating the mean vector of the classified spectral vectors according to the classes by using the final classification map p3Subtracting the mean vector of the class from the spectrum vector of each class to obtain the residual vector of the spectrum vectorWhereinRepresents the mean vector of all spectral vectors in class i, L-0, 1,2, …, L,the subscript i, j of (a) indicates the row and column of the spectrum vector corresponding to the residual vector in the image, the subscript l indicates the category of the residual vector, finally the residual vector is output to the spectrum transformation module, and the final classification map p3 and the mean vector corresponding to each category are outputThe side information is stored and transmitted to a decoding end;
(5) for residual error vectorPerforming a one-dimensional DCTAnd transforming to obtain a DCT (discrete cosine transformation) transformation coefficient, outputting the DCT transformation coefficient serving as a final transformation result to a coding unit of a three-dimensional multi-level tree set splitting algorithm 3DSPIHT to obtain a compressed code stream file with an accurately controllable code rate, and ending coding.
2. The method for lossy compression of hyperspectral images based on classified discrete cosine transform of claim 1, wherein the method of step (3a) is based on spectral vectorsAnd classifying the size of the mean value according to the following steps:
(3a1) finding each spectral vectorMean value ofWherein,representing the mean value of the spectral vectors of the ith row and the jth column in the mth group of spatial wavelet transform coefficients;
(3a2) to the mean value of the calculated spectral vectorsAnd (4) classifying:
first, the maximum variation range of the absolute values of all spectral vectors is determinedAnd dividing the range into a plurality of intervals by taking the integral power of 2 as a threshold:
th is a set of partitioned areas;
then, for the interval at 0Is characterized in that the variation range of the average value is very small, a mean value combining threshold Tm is selected, and the average value is [ -2 [)Tm,2Tm]The categories in between are combined into one category:th' is a set of combined intervals, and Tm is determined according to the gray value of the compressed image;
finally, according to the interval divided by Th', the cells in the same intervalAre classified into one category.
3. The method for lossy compression of hyperspectral images based on classified discrete cosine transform according to claim 1, wherein the step (3b) is performed by classifying the spectral vectors whose variation ranges are in the same interval into one class by using the variation ranges of the individual spectral vectors of each class in the initial classification map p1 according to the following steps:
(3b1) calculating the variation range of each spectral vectorWhereinRepresenting the variation range of the spectral vector at the ith row and jth column in the mth group, and finding out all the spectral vector variation ranges belonging to the same class according to the initial classification map p1Where L is 0,1,2, …, L1，L1Is the total number of categories in the initial classification map p 1;
(3b2) for the variation range of each spectral vector in class IAnd (4) classifying:
first, all the parameters are obtainedMaximum range of variation of absolute valueAnd dividing the range into a plurality of intervals by taking the integral power of 2 as a threshold:Th1is a collection of partitioned areas;
then, aiming at the characteristic that the variation range of the interval at 0 is very small, a variation range combination threshold Td is selected, and the range combination threshold Td is [ -2 ]Td,2Td]The categories in between are combined into one category:Th1' is a set of merged intervals;
finally, according to Th1' divided intervals, will be in the same intervalAre classified into one category.
4. The method for lossy hyperspectral image compression based on classified discrete cosine transform according to claim 1, wherein the step (3c) of performing the final classification by using the distribution positions of the maximum and minimum values of the spectral vectors of each class in the refined classification map p2 is performed according to the following steps:
(3c1) finding out all spectral vectors belonging to the same class according to the refined classification map p2, and finding out the positions of the maximum value component and the minimum value component of each spectral vector in the class in the spectral bands:andwherein k is1The spectral sequence number, k, representing the maximum component of the spectral vector2The sequence number, k, of the spectral band in which the minimum component of the spectral vector is located1And k2The following four cases are distributed:
<mfenced open = "{" close = "}"> <mtable> <mtr> <mtd> <mrow> <mn>1</mn> <mo>&le;</mo> <msub> <mi>k</mi> <mn>1</mn> </msub> <mo>&le;</mo> <mi>P</mi> <mo>/</mo> <mn>2</mn> <mo>,</mo> <mn>1</mn> <mo>&le;</mo> <msub> <mi>k</mi> <mn>2</mn> </msub> <mo>&le;</mo> <mi>P</mi> <mo>/</mo> <mn>2</mn> <mo>;</mo> </mrow> </mtd> <mtd> <mrow> <mi>P</mi> <mo>/</mo> <mn>2</mn> <mo>&le;</mo> <msub> <mi>k</mi> <mn>1</mn> </msub> <mo>&le;</mo> <mi>P</mi> <mo>,</mo> <mi>P</mi> <mo>/</mo> <mn>2</mn> <mo>&le;</mo> <msub> <mi>k</mi> <mn>2</mn> </msub> <mo>&le;</mo> <mi>P</mi> <mo>;</mo> </mrow> </mtd> </mtr> <mtr> <mtd> <mrow> <mn>1</mn> <mo>&le;</mo> <msub> <mi>k</mi> <mn>1</mn> </msub> <mo>&le;</mo> <mi>P</mi> <mo>/</mo> <mn>2</mn> <mo>,</mo> <mi>P</mi> <mo>/</mo> <mn>2</mn> <mo>&le;</mo> <msub> <mi>k</mi> <mn>2</mn> </msub> <mo>&le;</mo> <mi>P</mi> <mo>;</mo> </mrow> </mtd> <mtd> <mrow> <mi>P</mi> <mo>/</mo> <mn>2</mn> <mo>&le;</mo> <msub> <mi>k</mi> <mn>1</mn> </msub> <mo>&le;</mo> <mi>P</mi> <mo>,</mo> <mn>1</mn> <mo>&le;</mo> <msub> <mi>k</mi> <mn>2</mn> </msub> <mo>&le;</mo> <mi>P</mi> <mo>/</mo> <mn>2</mn> </mrow> </mtd> </mtr> </mtable> </mfenced>
(3c2) for all the spectral vectors in the same class obtained in the previous step, the sequence number k of the spectral segment where the maximum value component is located1And the spectral sequence number k of the minimum component2The spectral vectors of the same situation are divided into a class to obtain the final classification result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510097077.XA CN104639947B (en) | 2015-03-05 | 2015-03-05 | High spectrum image compression method based on classification discrete cosine transform |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510097077.XA CN104639947B (en) | 2015-03-05 | 2015-03-05 | High spectrum image compression method based on classification discrete cosine transform |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104639947A CN104639947A (en) | 2015-05-20 |
| CN104639947B true CN104639947B (en) | 2018-03-23 |
Family
ID=53218183
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510097077.XA Active CN104639947B (en) | 2015-03-05 | 2015-03-05 | High spectrum image compression method based on classification discrete cosine transform |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104639947B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107835429B (en) * | 2017-10-31 | 2020-10-13 | 北京信息科技大学 | Hyperspectral image compression method and system |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6535647B1 (en) * | 1999-03-25 | 2003-03-18 | General Dynamics Decision Systems, Inc. | Image encoder method |
| CN103391438A (en) * | 2013-07-19 | 2013-11-13 | 哈尔滨工程大学 | Hyper-spectral image compression and encoding method and device |
| CN103501437A (en) * | 2013-09-29 | 2014-01-08 | 北京航空航天大学 | Fractal and H.264-based hyper-spectral image compression method |
| CN104156954A (en) * | 2014-08-01 | 2014-11-19 | 西安电子科技大学 | Registration pre-processing system suitable for multispectral image compression |
-
2015
- 2015-03-05 CN CN201510097077.XA patent/CN104639947B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6535647B1 (en) * | 1999-03-25 | 2003-03-18 | General Dynamics Decision Systems, Inc. | Image encoder method |
| CN103391438A (en) * | 2013-07-19 | 2013-11-13 | 哈尔滨工程大学 | Hyper-spectral image compression and encoding method and device |
| CN103501437A (en) * | 2013-09-29 | 2014-01-08 | 北京航空航天大学 | Fractal and H.264-based hyper-spectral image compression method |
| CN104156954A (en) * | 2014-08-01 | 2014-11-19 | 西安电子科技大学 | Registration pre-processing system suitable for multispectral image compression |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104639947A (en) | 2015-05-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107392952B (en) | no-reference mixed distortion image quality evaluation method | |
| CN105719322A (en) | Multispectral image compression method based on square matrix transformation | |
| Karami et al. | Hyperspectral image compression using 3D discrete cosine transform and support vector machine learning | |
| CN107133992A (en) | Based on spectrum correlation adaptive grouping high spectrum image distributed associating compression method | |
| CN111598786A (en) | Hyperspectral image unmixing method based on deep denoising self-coding network | |
| Shi et al. | A novel vision-based adaptive scanning for the compression of remote sensing images | |
| CN103297775B (en) | Based on the Multispectral Image Compression System and method for that unicode rate distributes and controls | |
| JP3828640B2 (en) | Image signal conversion method | |
| Hassanzadeh et al. | Compression and noise reduction of hyperspectral images using non-negative tensor decomposition and compressed sensing | |
| CN104639947B (en) | High spectrum image compression method based on classification discrete cosine transform | |
| CN106331719B (en) | A kind of image data compression method split based on the Karhunen-Loeve transformation error space | |
| Sadkhan | A proposed image compression technique based on DWT and predictive techniques | |
| Baviskar et al. | Performance evaluation of high quality image compression techniques | |
| Can et al. | Compression of hyperspectral images using luminance transform and 3D-DCT | |
| Naimi et al. | Multiple description image coding using contourlet transform | |
| Kozhemiakin et al. | An approach to prediction and providing of compression ratio for DCT-based coder applied to multichannel remote sensing data | |
| Baviskar et al. | Performance analysis of sub-band replacement DWT based image compression technique | |
| Kaarna et al. | Compression and classification methods for hyperspectral images | |
| Narmadha et al. | An optimal HSI image compression using DWT and CP | |
| Keranen et al. | Spectral similarity measures for classification in lossy compression of hyperspectral images | |
| Li et al. | Efficient multi-bands image compression method for remote cameras | |
| Ponomarenko et al. | Lossy and visually lossless compression of single-look SAR images | |
| Zhu | On the performance of JPEG2000 and principal component analysis in hyperspectral image compression | |
| Zhou et al. | Satellite hyperspectral imagery compression algorithm based on adaptive band regrouping | |
| Mishra et al. | A hybrid scheme of image compression employing wavelets and 2D-PCA |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |