CN1595444A - Ultra-spectrum image real-time compression system based on noise decomposing compression algorithm - Google Patents

Abstract

The invention relates to a hyperspectrum image real-time compression system based on the noise decomposition and compression algorithm, which is composed of the host computer, the high-speed image processor and the image collection apparatus. The image collection apparatus is connected to the slave processor via the CAB bus port and to the host computer via the PCI bus. The host computer is bi-directionally connected to the slave processor via the PCI bus. In addition, it is connected to the image display and the image data compression result memory respectively. Its characteristic lies in the following. First, the bi-directionally connected image noise decomposition and compression block and the host connector is equipped on the slave processor to receive the command data from the host computer and to transmit the image real time compression result data to the host computer. Second, the host computer control block, the slave processor interface, the display block and the storage block are equipped on the host computer. The control block receives the original image data and the user command to generate the system data resource, and transmits the command to the slave processor and receives the image real time compression result data via the slave processor. The host computer control block transmits the original image data to the display block and the image real time compression result to the storage block to generate the compression file.

Description

HYPERSPECTRAL IMAGERY real-time compression system based on noise decomposition pressure compression algorithm

Technical field

The present invention relates to a kind of HYPERSPECTRAL IMAGERY real-time compression system, particularly a kind of image compression system based on noise decomposition pressure compression algorithm, signal-to-noise ratio (snr) estimation by real-time HYPERSPECTRAL IMAGERY, the noise that to be introduced by compression algorithm is reduced to minimum to the influence of image information, realized high-fidelity compression, also guaranteed the image compression system of compression of images rate simultaneously HYPERSPECTRAL IMAGERY.

Background technology

To the existing many reports of the achievement in research of HYPERSPECTRAL IMAGERY compression, but rest on theoretical research and software processes stage mostly, accomplish the few of real-time hardware adaptations.Trace it to its cause, the one, the hyperspectral image data amount is huge, and is very high to the hardware handles Capability Requirement, is difficult in the past finish in real time with hardware circuit; The 2nd, HYPERSPECTRAL IMAGERY to be carried out the application often of Compression Study and handle unit, its research work does not have and the development work of imaging spectrometer is closely linked, seldom by practical application.

Along with the development of large scale integrated circuit technology, many high speeds, high reliability, the integrated circuit (IC) chip of low-power consumption is developed out, and the real-time processing of the hardware of digital picture has become possibility.Utilize multi-disc high-performance treatments chip to form parallel processing system (PPS), the real-time compression system that constitutes in conjunction with the HYPERSPECTRAL IMAGERY compression algorithm can reach the practical stage.And the miniaturization of imaging spectrometer requires and the pressure of a large amount of hyperspectral image data storage and transmission also makes the development department of imaging spectrometer begin to pay attention to the research work of this respect.The coastal waters imager (COIS) that carries as the NEMO naval map observation satellite by the United States Naval Research Laboratory research and design comprises two imagers, comprises that respectively 60 and 150 width are the spectral band of 10nm.The data transfer rate of these two imagers has reached 145Mb/s.In order to reach 10 times data compression rate (diminishing), alleviate that HYPERSPECTRAL IMAGERY is stored and the burden when transmitting earthward on star, United States Naval Research Laboratory has been developed real time processing system on a feature extraction and the data compression star, is called optics real-time adaptive spectrum identification system (ORASIS).In the experiment that unmanned spacecraft carries out, the actual compression effect that ORASIS reaches is: when ratio of compression=12, and Y-PSNR=48.0db; When ratio of compression=20, Y-PSNR=44.2db.In China, relevant research was also done by units such as BJ University of Aeronautics ﹠ Astronautics, the National University of Defense technology, Tsing-Hua University, Harbin Institute of Technology.Form by green (0.52-0.6 μ m), red (0.63-0.69 μ m), three wave bands of near infrared (0.76-0.94 μ m) by the imaging spectrometer that " No. one, space flight Tsing-Hua University " microsatellite of Chinese Tsing-Hua University, company of China Aerospace Machinery and Electronics Corporation and Britain Sa Rui university joint research and development carries, adopt a-MPBTC (Adaptive Moment PreservingBlock Truncation Coding) compression method, obtain 4 times compressibility.

But all there is the interference of random noise inevitably in actual remote sensing images, analyze these imaging spectrum compressibilities, its weak point is the noisiness that they do not consider image, when image is carried out the data compression of certain multiple, the original noise of image border and transitional zone is amplified by system, and spread, polluteed the higher bit data of image, caused the loss of image information.When utilizing these compressibilities that image is compressed in brief, image information can not reasonably be kept.

Summary of the invention

As mentioned above, how according to the noise pollution situation of HYPERSPECTRAL IMAGERY, the compression multiple that improves image under the condition that does not increase image information loss is a technical matters to be solved by this invention.Therefore, the object of the present invention is to provide a kind of HYPERSPECTRAL IMAGERY compressibility, by noise analysis to each wave band data of original image, design effective method for compressing image, each wave band data of HYPERSPECTRAL IMAGERY is compressed targetedly, set up HYPERSPECTRAL IMAGERY real-time compression system based on noise decomposition pressure compression algorithm.

Technical solution of the present invention is as follows:

According to a kind of HYPERSPECTRAL IMAGERY real-time compression system of the present invention based on noise decomposition pressure compression algorithm, comprise main frame, high speed image from processor and image collecting device, described image collecting device is connected and is connected with main frame through pci bus from processor with high speed image through the CAB bus interface; This main frame through pci bus with become double-way connection from the high speed image processor, and also connect image display and Image Data Compression result memory respectively, characteristics are:

A. be provided with into picture noise decomposition compression module and the host interface that bidirectional traffic connects at described high speed image from processor, send main frame to from the order data of main frame with image Real Time Compression result data so that accept;

B. on main frame, be provided with host computer control module, from processor interface and display module and memory module, this control module is accepted the order that raw image data that image collecting device sends here and main frame send here and is formed the data source of system, and through sending order and accept the image Real Time Compression result data of force feed from processor to high speed image from processor interface; Host computer control module is sent raw image data and is shown mould and send memory module formation compressed file to leave in the Image Data Compression result memory image Real Time Compression result data.

Further, described main frame is portable PC;

Described high speed image is made of a plurality of dsp chip image processing boards from processor;

Described picture noise decomposes compression module and comprises the original image input block that connects successively, the signal noise ratio (snr) of image evaluation unit, noise decompose quantizer and side by side after connect lossless compression-encoding unit and the lossy compression method coding unit that this noise decomposes quantizer, and the lossless compress decoding unit that connects this lossless compression-encoding unit with Channel Transmission after respectively with after connect the noise compositor of the lossy compression method decoding unit of this lossy compression method coding unit and be connected in parallel this lossless compression-encoding unit and lossy compression method decoding unit;

Described lossless compression-encoding unit comprises that the image subblock cutting unit, the image subblock two dimension that connect with data stream successively quantize cosine transform unit and coded image data compression unit, and wherein said image subblock cutting unit is divided into 8 * 8 with original image;

Described lossy compression method coding unit comprises that the image subblock cutting unit, the image subblock two dimension that connect with data stream successively quantize cosine transform unit, quantifying unit and coding unit, and wherein said image subblock cutting unit is divided into 8 * 8 sub-pieces with original image.

Advantage of the present invention is:

Because the present invention has carried out the separation of bit field to image according to the noise size, and height bit field data are taked the strategy that compresses respectively, make the noise limit introduced by compression algorithm that the low bit field of noise pollution arranged, can not avoid the secondary pollution of noise to the diffusion of higher bit numeric field data to image information.

Put it briefly, the present invention has made full use of the noisiness of HYPERSPECTRAL IMAGERY, the noise that to be introduced by compression algorithm is reduced to minimum to the influence of image information, realized high-fidelity compression to HYPERSPECTRAL IMAGERY, while has also guaranteed the compression of images rate, compares with traditional HYPERSPECTRAL IMAGERY compressibility to have tangible technical progress and substantive meaning.

Description of drawings

Fig. 1 is a hardware architecture synoptic diagram of the present invention.

Fig. 2 is a software architecture synoptic diagram of the present invention.

Fig. 3 is a multi-disc DSP pipelined-flash processing mode synoptic diagram on the image processing board of the present invention.

Fig. 4 is a monolithic DSP two-stage double buffering streamline workflow synoptic diagram of the present invention.

Fig. 5-the 1st, the structural representation of noise decomposition pressure compression algorithm of the present invention.

Fig. 5-the 2nd, the process flow diagram based on noise decomposition pressure compression algorithm of the present invention.

Fig. 5-the 3rd, the structural representation based on the lossless compression-encoding unit in the noise decomposition pressure compression algorithm of the present invention.

Fig. 5-the 4th, the structural representation based on the lossy compression method coding unit in the noise decomposition pressure compression algorithm of the present invention.

Embodiment:

Provide better embodiment of the present invention according to Fig. 1～Fig. 5-4 below, and described in detail,, rather than be used for limiting protection scope of the present invention so that architectural feature of the present invention and function characteristics are described better.

Consult Fig. 1, hardware architecture of the present invention comprises that system's main control system 10, high speed image are from processor 20, image acquisition board 22, CAB bus interface 21 etc.Wherein: system's main control system 10 adopts portable PC, the image processing board Python/C6 disposable plates that high speed image adopts based on multi-disc TMS320C6201 from processor 20.Described main frame 10 tasks are master control total systems, utilize its friendly interface to receive user command, start/stop from the work of processor 20 (being that high speed image is from processor 20), working routine of change DSP on the disposable plates etc., and the view data of gathering and the relevant information of compression of images shown in real time, the data storage after the compression is got off.In addition, after the software of main frame 10 also is responsible for the real-time working end, the decompress(ion) playback of compression result work during aftertreatment.The task of Python/C6 disposable plates then is to accept view data, and the program that allows DSP select according to main frame is handled, the output compression result.The original spectrum view data that described image acquisition board 22 is gathered is divided the two-way transmission, and the CAB bus interface 21 of leading up to is directly inputted among the Python/C6 to be handled; The computer PCI bus 30 of leading up to is sent to main frame 10 to carry out Real Time Compression and by image display 12 demonstrations, is convenient to monitoring.The data compression result of Python/C6 then passes to main frame 10 by pci bus 30, by storer 11 storages or further processing.

Consult Fig. 2, software architecture of the present invention is divided into main frame 10 platforms (Host) and from processor 20 platforms (Slave) two parts.Main modular on the main frame 10 has main control module 101, from processor interface module 102, image display 103 and data memory module 104.Its major function provides good man computer interface, accepts user command, and control receives the compression result of microprocessor from the operation of processor 20, utilizes the real-time demonstration of technology realization original images such as multithreading and the memory function of compression result.From the main modular of processor 20 softwares image compression algorithm module 201 and host interface module 202 are arranged.Its major function then is to utilize multiple code optimization means on the DSP platform, and uses the high-level efficiency code encoder of TI company, gives full play to the hardware resource performance of DSP, finishes the data in real time compression work and passes to main frame according to user's order.Whole software with original image and user command as data source, with the compression result end of file.

The HYPERSPECTRAL IMAGERY compressibility of present embodiment is characterized in: for designed HYPERSPECTRAL IMAGERY compression algorithm, can not add modification or few as far as possible ground of revising with the hardware system realization, save the required workload of algorithm of reappraising.

The HYPERSPECTRAL IMAGERY compressibility of present embodiment is characterized in: stronger dirigibility is arranged, and system can easily carry out the improvement of algorithm at application request, and the whole hardware system of development and Design again.

The HYPERSPECTRAL IMAGERY compressibility of present embodiment is characterized in: have extensibility and portability, hardware design is convenient to system and is expanded later on or be applied to new platform.

The compression based on noise branch HYPERSPECTRAL IMAGERY in the present embodiment is based on the compression algorithm that noise decomposes, and its step is included in: S1. obtains view data from image pick-up card; S2. the signal to noise ratio (S/N ratio) of HYPERSPECTRAL IMAGERY is estimated; S3. view data is carried out physical separation in " bit field ", be divided into clean higher bit numeric field data and the low bit field data that noise pollution is arranged; S4. higher bit area image data are carried out (standard) lossless compress, the fidelity image information is carried out lossy compression method to low bit field view data to greatest extent, increases the compression of images rate; S5. compression result is delivered to host stores.

Be illustrated in figure 3 as multi-disc DSP pipelined-flash processing mode synoptic diagram on the Python/C6 image processing board of the present invention.Every DSP individual processing one circuit-switched data of parallel processing mode is finished a complete compression algorithm task, after each circuit-switched data all disposes, all modules is exported simultaneously.The advantage of this structure is that Task Distribution is simple, and each DSP function, algorithm are identical, is easy to realize.In system of the present invention, characteristics in conjunction with many DSP image processing board, adopted the method for parallel processing of timesharing flowing water, that is: with the hyperspectral image data of importing, according to the size of the memory buffer of each DSP configuration, be divided into a certain size frame, flowing water is given each DSP in chronological order, and the result of each DSP also combines in chronological order, is transferred to main frame 10 by pci bus.

Selected DSP is the TMS320C6201 chip of TI company in the present embodiment, and CPU can visit IDRAM twice in a clock period, and comparing will be through the external memorizer of EMIF extended memory interface and external bus, and efficient is much higher.In order to give full play to the performance of DSP, will be the buffer zone of IDRAM as exchanges data.But for the compression of images of needs processing mass data, the IDRAM capacity is too little, can not satisfy the needs of rapid data input and output.For this reason, adopted the two-level cache structure: chip external memory SDRAM (16MB) is as frame buffer, and IDRAM (64KB) is as the image subblock buffer memory.Raw image data earlier input is temporarily stored in the frame buffer, is that unit sends in the sheet block cache and handles to CPU with sub-piece again.The dma controller of C6201 has the passage of 4 separate programmings, allows the DMA that carries out 4 different contents to transmit.For the intensive image compression system of storage, the I/O handling capacity is most important to the processing power and the efficient of system.Utilize Python/C6 to go up four DMA of C6201, can realize two-stage I/O buffering, i.e. the exchange of view data relatively slow SDRAM storer from the bus to C6201, and the SDRAM storer is to the exchange of the interior high-speed data RAM of sheet of C6201.Like this, data compression process can be carried out among the high-speed data RAM in sheet, and the handling capacity of data is greatly improved.

Be illustrated in figure 4 as monolithic DSP two-stage double buffering streamline workflow synoptic diagram.Realize of the operation of first order input equipment respectively with DMA2 and DMA3 to incoming frame buffer memory BufFrameIn and output frame buffer memory BufFrameOut to output device, with DMA0 and DMA1 realize respectively second level incoming frame buffer memory BufFrameIn in the sheet in sub-block cache BufBlockIn and the sheet block cache BufBlockOut operate to output frame buffer memory BufFrameOut.This two-stage has all been utilized ping-pong buffer, and the method for using pipeline parallel method to handle makes input and output and data compression process carry out simultaneously.Dotted line among the figure and solid line are respectively the operation of two mutual exclusions, can not carry out simultaneously, but can carry out simultaneously with the operation that a kind of line is represented.The operation that dotted line and solid line are represented is alternately carried out and is formed streamline.Concrete pipeline parallel method processing procedure following (establishing two level production line initialization filling finishes):

1) starts DMA2, gather the next frame view data to BufFrameIn1 from input equipment;

2) start DMA3, the result outputs to pci bus from BufFrameOut1 with the previous frame compression of images;

3) start DMA0, next image subblock among this frame image data BufFrameIn2 is sent to BufBlockIn2;

4) start DMA1, will go up an image subblock compression result and be sent to BufFrameOut2 from BufBlockOut2;

5) this image subblock among the compression treatments B ufBlockIn1, the result puts into BufBlockOut1;

6) wait transmits end up to DMA0, DMA1, and repeating step 3,4,5 also is used alternatingly sub-block cache in two sheets, all compresses up to this frame to finish;

7) wait for and to transmit end that repeating step 1,2,3,4,5,6 also is used alternatingly and extends out frame buffer, begins to compress subsequent frame up to DMA2, DMA3.

Such processing procedure makes the transmission of 4 DMA channel datas and DSP kernel compression work parallel fully, and has guaranteed that main calculation task carries out in sheet is internally cached, has given full play to the performance of system hardware, has improved data throughput capabilities.

Shown in Fig. 5-1 and Fig. 5-2, the HYPERSPECTRAL IMAGERY compression in the present embodiment has made full use of the noisiness of image, before image is compressed, at first will estimate the signal to noise ratio (S/N ratio) of image.Present embodiment to the basic thought of signal noise ratio (snr) of image estimation is: owing to select a certain size homogeneous area relatively more difficult, just segment the image into zonule one by one, can think basically in these zonules uniformly; Calculate LSD (Local Standard Deviation, local standard is poor) in these zonules respectively as the local noise size, and select the average noise of that maximum interval LSD of sum as entire image.Concrete operation steps is as follows:

Image segmentation is become 4 * 4, or 5 * 5 ..., or 8 * 8 fritter, for each image subblock, the LM of signal (Local Mean local mean value) is obtained by following formula:

$\mathrm{LM}=\frac{1}{N}\underset{i=1}{\overset{i=N}{\mathrm{\Σ}}}{S}_i$

Here, Si is the gray-scale value of i pixel in the image subblock; N is the sum of all pixels in the image subblock.LSD is obtained by following formula:

$\mathrm{LSD}={\left[\frac{1}{N-1}\underset{i=1}{\overset{i=N}{\mathrm{\Σ}}}{(S_i-\mathrm{LM})}^2\right]}^{1/2}$

For the sub-piece of uniform image, LSD is less, and to uneven image subblock, as comprises the sub-piece of image border or textural characteristics, and LSD is then bigger.Calculate the LM (being designated as LMo) of entire image, the LSD of the sub-piece of all images, and find out minimum and maximum LSD in the sub-piece of all images.

Between minimum and maximum LSD, set up several equivalent intervals at interval.The LSD of all sub-pieces is entered corresponding interval successively according to the size of value.Number to each interval LSD is counted, and the mean value of that interval LSD of count value maximum is the noise of entire image, is designated as LSDo.

Can try to achieve the signal to noise ratio snr of entire image by following formula:

$\mathrm{SNR}=20\log \frac{{\mathrm{LM}}_o}{{\mathrm{LSD}}_o}$

Be depicted as the structural representation that decomposes the image compression module of compression based on noise as Fig. 5-1.The basic thought of realizing this compression is: at first, raw data is quantized, then quantized signal and error signal are done the compression processing respectively.For quantized signal, adopt (standard) lossless compression algorithm, carry out information fidelity storage to greatest extent, for quantizing noise (error) signal, adopt lossy compression comparatively efficiently.

Decompose in the quantizer at noise, make and get:

S _H=[S/q] S _L=S-S _HQ [*] is for rounding operation

In the noise compositor, get:

$\hat{S}={\hat{S}}_H\·q+{\hat{S}}_L\mathrm{mod}\left(q\right)$ Mod (*) is a congruence expression

If adopt undistorted coding, make ${\hat{S}}_H=S_H$

Can obtain so:

$[\hat{S}/q]={\hat{S}}_H=S_H=[S/q]$

That is to say, no matter to S _LAdopt what kind of compression scheme, always can guarantee that all decompressed datas are believable at last on the following formula meaning.The thought essence of the compression algorithm of decomposing based on noise is: the isolation of signal having been carried out physics on " bit field " is divided, congruence operation by compositor, make the strictness of compression algorithm noise be limited in low bit field, the diffusion to the higher bit territory can not take place.

It is different carrying out " bit field " separation for the view data of different original image signal to noise ratio (S/N ratio)s, signal to noise ratio (S/N ratio) is high more, then the bit of higher bit territory distribution is many more, the bit that low bit field is distributed is few more, on the contrary, signal to noise ratio (S/N ratio) is low more, and then the bit of higher bit territory distribution is few more, and the bit that low bit field is distributed is many more.As following table:

The video bits territory breakdown of noise decomposition pressure compression method

Original image SNR (db)	Original image signal noise amplitude ratio	Bit field is decomposed
			????≥54db	????512∶1	High 9bit hangs down 3bit
????≥48db	????256∶1	High 8bit hangs down 4bit
			????≥42db	????128∶1	High 7bit hangs down 5bit
????≥36db	????64∶1	High 6bit hangs down 6bit
			????≥30db	????32∶1	High 5bit hangs down 7bit
????≥24db	????16∶1	High 4bit hangs down 8bit
			????≥18db	????8∶1	High 3bit hangs down 9bit

By means of the SNR estimation result of HYPERSPECTRAL IMAGERY, the bit field that can obtain each band image of HYPERSPECTRAL IMAGERY is decomposed, and then realizes corresponding (standard) lossless compress and high power compression.Like this, originally clean bit remains clean, originally had the bit of noise pollution that noise is still arranged, the noise that lossy compression method algorithm itself is brought can only continue in the low bit field of original noise pollution to pollute, and can not pollute the data in higher bit territory.

Press Fig. 5-1 to shown in Fig. 5-4, the basic step of summarize is: step 51 is input to original image input block 2010 to original image from image acquisition board 22; Step 52 is estimated by 2011 pairs of input picture signal to noise ratio (S/N ratio)s of signal noise ratio (snr) of image evaluation unit; Step gathers 53, decomposes quantizer by picture noise image is carried out the bit field separation according to the signal-to-noise ratio (snr) estimation result, is divided into higher bit area image and low bit field image; Step 55 is carried out (standard) lossless compress by the 2013 pairs of higher bit area images in lossless compression-encoding unit; Step 551 is divided into 8 * 8 image subblocks one by one by image subblock cutting unit 20131 with image; Step 552 quantizes the discrete cosine transform (2DCT) that 20132 pairs of each image subblocks in cosine transform unit carry out two dimension by the image subblock two dimension; Step 553 is directly carried out the Huffman coding by 20133 pairs of transformation results of coded image data compression unit without quantification, realize the high-fidelity low range compression to image; Step 54; Carry out lossy compression method by 2014 pairs of low bit field images of lossy compression method coding unit; Step 541 is divided into 8 * 8 image subblocks one by one by image subblock cutting unit 20141 with image; Step 542 quantizes the discrete cosine transform (2DCT) that 201412 pairs of each image subblocks in cosine transform unit carry out two dimension by the image subblock two dimension; Step 543 is quantized by 20143 pairs of transformation results of quantifying unit; Step 544, carry out the Huffman coding by data after 20144 pairs of quantifications of coding unit, realization is compressed the high magnification that diminishes of image, at last, step 56 is passed through Channel Transmission, the low letter rate compress coding data of the image high-fidelity in the lossless compression-encoding unit 2013 is sent to lossless compress decoding unit 2015 and the lossy compression method coded data is sent to lossy compression method decoding unit 2016 also send into noise compositor 2017 respectively.

Claims (8)

1, a kind of HYPERSPECTRAL IMAGERY real-time compression system based on noise decomposition pressure compression algorithm, comprise that main frame (10), high speed image are from processor (20), and image collecting device (22), this image collecting device (22) through CAB bus interface (21) be connected from processor (20) and be connected with main frame (10) through pci bus (30); This main frame (10) through pci bus (30) with become double-way connection from processor (20), and this main frame (10) connects image display (12) and Image Data Compression result memory (11) respectively, it is characterized in that:

A. decompose compression module (201) and host interface (202) at the picture noise that is provided with into double-way connection from processor (20), send main frame (10) to from the order data of main frame (10) with image Real Time Compression result data so that accept;

B. on main frame (10), be provided with host computer control template (101), from processor interface (102) and display module (103) and memory module (104), this control module (101) is accepted raw image data and user command and is formed DSN, and through from processor interface (102) to sending order and acceptance pattern from processor (20) as the Real Time Compression result data; Host computer control module (101) is sent raw image data display module (103) and is sent memory module (104) to form compressed file image Real Time Compression result.

2, the HYPERSPECTRAL IMAGERY real-time compression system based on noise decomposition pressure compression algorithm according to claim 1 is characterized in that: described main frame (10) is portable PC.

3, the HYPERSPECTRAL IMAGERY real-time compression system based on noise decomposition pressure compression algorithm according to claim 1 is characterized in that: describedly be made of multi-disc dsp chip image processing board from processor (20).

4, according to claim 1 or 3 described HYPERSPECTRAL IMAGERY real-time compression systems based on noise decomposition pressure compression algorithm, it is characterized in that: described picture noise decomposes compression module (201) and comprises the original image input block (2010) that connects successively, signal noise ratio (snr) of image evaluation unit (2011), noise decomposes quantizer (2012), with side by side after connect the lossless compression-encoding unit (2013) and lossy compression method coding unit (2014) that this noise decomposes quantizer (2012), and the lossless compress decoding unit (2015) that connects this lossless compression-encoding unit (2013) with Channel Transmission after respectively with after connect diminishing of this lossy compression method coding unit (2014) and press decoding unit (2016) and this lossless compress decoding unit (2015) that is connected in parallel, the noise compositor (2017) of lossy compression method decoding unit (2016).

5, the HYPERSPECTRAL IMAGERY real-time compression system based on noise decomposition pressure compression algorithm according to claim 4 is characterized in that: described lossless compression-encoding unit (2013) comprises that the image subblock cutting unit, the image subblock two dimension that connect with data stream successively quantize cosine transform unit and coded image data compression unit.

6, the HYPERSPECTRAL IMAGERY real-time compression system based on noise decomposition pressure compression algorithm according to claim 5, it is characterized in that: described image subblock cutting unit is divided into 8 * 8 sub-pieces to original image.

7, the HYPERSPECTRAL IMAGERY real-time compression system based on noise decomposition pressure compression algorithm according to claim 4 is characterized in that: described lossy compression method coding unit (2014) comprises that the image subblock cutting unit, the image subblock two dimension that connect with data stream successively quantize cosine transform unit, quantifying unit and coding unit.

8, the HYPERSPECTRAL IMAGERY real-time compression system based on noise decomposition pressure compression algorithm according to claim 7, it is characterized in that: described image subblock cutting unit is divided into 8 * 8 sub-pieces to original image.

Images