CN103747267A - Compressive sensing spectral imaging measured value compression method based on embedded transformation coding - Google Patents

Abstract

The present invention proposes a compression sensing spectral imaging measurement value compression method based on embedded transformation coding, which converts the original sampling measurement value matrix into two matrices by using the transformation coding idea, and then compresses the two matrices separately, which is specifically divided into two Two stages: In the first stage, the original sampling measurement value is converted into an image matrix that is more similar to the distribution of the original spectral channel image and a random and sparse remainder matrix by using the mean value filter transformation based on the coded aperture information; the second stage In , the positions where the element bits are zero and the bit planes are zero in the remainder matrix are determined by encoding the aperture information, and the bit plane encoding is used to efficiently compress the remainder matrix. The scheme converts the measurement value matrix, which is difficult to compress due to the approximate random combination, into two matrices with regular distribution and easy processing through embedded transformation coding, which effectively improves the lossless compression efficiency of the measurement values based on the coded aperture spectral imaging system.

Description

Compressed sensing light spectrum image-forming measured value compression method based on embedded transition coding

Technical field

The present invention relates to a kind of information compressing method towards code aperture snapshot spectrum imaging system sampled measurement, thus a kind of method of utilizing known code aperture information to carry out transition coding and Bit-Plane Encoding the sampled measurement of code aperture snapshot spectrum imaging system is carried out Lossless Compression specifically.

Background technology

Compressed sensing is take the compressibility of signal or sparse property as condition, it is a brand-new signal processing theory that breaks through traditional nyquist sampling theorem, it makes us can in signal sampling, complete compression coding, and the development of signal processing is had great significance.Compressed sensing theory reaches its maturity, and scholars have launched application study widely in various fields to it.Multi-optical spectrum imaging technology is one of them important branch.Light spectrum image-forming is as a kind of emerging technology, and it can produce the spatial distribution map of spectrum change, makes it in many application, become very powerful and exceedingly arrogant instrument.The signal that traditional sampling pattern gathers is redundancy, so hits can compress, and compressed sensing requires to provide the sensing matrix (sampling matrix) of incoherent sampling, the incoherence of sampling is exactly that the data that collect with sensing matrix of requirement should not be present in sparse base, and the sampled value of compressing like this could be preserved more information as much as possible.But correspondingly, also redundancy no longer of the final measured value signal that sensing matrix obtains, and be not easy to compression.But in multi-optical spectrum imaging system, researchers still wish further to compress the sampled measurement after compressed sensing, so that its real-time Transmission when the application of the aspect such as environmental remote sensing, astrophysics and military target detection.

Through the literature search of prior art is found, recognize that the measured value data sum of most of light spectrum image-forming technology generations is more than or equal to the sum of measured value in reconstruction of three-dimensional data block, D.J.Brady in 2006 and M.E.Gehm have introduced compressible light spectrum image-forming thought on " the Compressive imaging spectrometers using coded apertures " of International Society for Optics and Photonics meeting, and the measured value that the method is intended to light spectrum image-forming is produced is less than the elements are contained value in reconstructed data block.Its theoretical foundation depends on compressive sensing theory, and the fact that can rarefaction representation on some base according to natural scene solves that data stereo block rebuilds owes to determine problem.They have carried out CAI design according to compressed sensing thought, studies have shown that and utilize this pseudorandom building method design code aperture can make observation model meet constraint equidistant characteristics (RIP) character.Subsequently, it (is CASSI that the people such as M.E.Gehm have delivered a kind of binary distributing code aperture snapshot spectrum imaging system in 2007 on " the Single-shot compressive spectral imaging with a dual-disperser architecture " at Optics Express periodical, the Coded Aperture Snapshot Spectral Imager), utilize the light field of code aperture and dispersive medium modulation scene, and utilize a detector to obtain two-dimensional representation and the multiplex projection of 3D data volume.The people such as A.Wagadarikar have reported a compressible CASSI system at " the Single disperser design for coded aperture snapshot spectral imaging " of Applied Optics periodical in 2008, are referred to as monochromatic loose CASSI instrument.Because the loose CASSI system of monochrome has been used frequency multiplexing technique, and optics is less, and this system is easy to registration, is applicable to the application demand of low spatial resolution and high spectral resolution.H.Arguello and G.R.Arce have proposed a broad sense and effective mathematical framework and corresponding coding aperture optimal method in 2013 " Rank minimization code aperture design for spectrally selective compressive imaging " at IEEE Trans.Image Process periodical is upper based on the monochromatic loose CASSI system of many bats (multi-shot), the method allows the reconstruction of any band subset, reduces to greatest extent required shooting number (shots) simultaneously.Because the recovery of sparse signal in compressed sensing is very responsive to compressible data and compression projection, different shooting used the Y-PSNR (PSNR) of the reconstruction image that different coding aperture obtains all lower under ways, although also rise gradually along with umber of beats increases PSNR, corresponding compression efficiency is also declining gradually.For limited samples size of data and guarantee the quality of data reconstruction, a solution is to provide the further Lossless Compression to sampled measurement, thereby under similar memory space, can utilize more shooting ways to improve PSNR.Relevance and redundancy in the measured value obtaining due to compressed sensing light spectrum image-forming are weakened greatly, and the compression duty of sampled measurement is for the beyond doubt individual new challenge of existing Lossless Compression research.

Summary of the invention

The present invention is directed to the not squeezable present situation of crude sampling measured value, proposed a kind of transform coding method based on mean filter conversion, by the effective utilization to Given information and system configuration, improved the compression efficiency of method.The method is by utilizing known code aperture information, process one by one the element in compressible measurement matrix, finally be converted into one and distribute and be similar to and be easier to the matrix that compresses with original image, and the sparse matrix of Bit-Plane Encoding (bit plane coding) is carried out in available code aperture as supplementary.

The present invention is achieved by the following technical solutions:

Original three-dimensional data is after the code aperture of CASSI system, dispersion element, detector, the compressible sampled measurement obtaining be actually different spectrum channels be encoded and translation after pixel sum, for the synthetic matrix of this random groups within the specific limits, it is two matrixes by former sampled measurement matrix conversion that the present invention utilizes transition coding thought, then two matrixes is compressed separately to processing; By two megastages, it carried out to splitting and reorganizing and choose the algorithm adapting and compress.In first stage, utilize the mean filter conversion based on code aperture information to convert former sampled measurement to an image array being more similar to the distribution of former spectrum channel image, and a random and sparse remainder matrix.In second stage, by code aperture information, determine that in remainder matrix, element bit is that zero-sum bit plane is zero position, and utilize Bit-Plane Encoding Efficient Compression remainder matrix.

Preferably, in the described first stage, in the mean filter conversion based on code aperture information, make full use of known code aperture information, be specially: according to the individual random coded aperture matrix that uses in system generation sampled measurement process, the addend number of each element in computation and measurement value matrix, the transformation idea of utilization based on mean filter by each element in measured value divided by its addend number, obtain an image array approximate with the distribution of former spectrum channel image, and a random and sparse remainder matrix.

Preferably, the image array that the described first stage changes out, the distribution of this matrix image and statistical nature and former spectrum channel image are similar, adopt Lossless Image Compression algorithm JPEG-LS and Calic directly to compress it.

Preferably, in described second stage, describedly by code aperture information, determine that in remainder matrix, element bit is that zero-sum bit plane is zero position, be specially: the each element in remainder matrix has some bits to be defined as zero according to code aperture information, these are defined as zero bit and need not list in the sequence of compression and go, and directly from decoding end, by code aperture information, recover; And uncertain be that zero bit will be processed successively by the Bit-Plane Encoding of bit-by-bit plane piecemeal portions.

Preferably, in described second stage, the described Bit-Plane Encoding Efficient Compression remainder matrix that utilizes, be specially: the method that adopts the Bit-Plane Encoding based on code aperture information for the remainder matrix after conversion, according to similar positional structure and statistical property piecemeal by stages adopt Bit-Plane Encoding and variable-length encoding to compress, block-by-block coding, each bit plane from the most remarkable bit plane to gradual ground of least remarkable bit plane code coefficient amplitude, on each bit plane, from part 1, to part 3, progressively encode, it is one group that each part is once only got four positions, because some position has been defined as zero according to code aperture information, therefore four bits are at most only contained in each group, finally these bit groups are compressed with variable-length encoding.

Compared with traditional Lossless Image Compression method, the CASSI systematic survey value compression method based on code aperture and embedded transition coding proposed by the invention, has improved further possibility and the practicality of compression of compressed sensing sampled measurement.The invention has the beneficial effects as follows: utilize the conversion of mean filter formula to be converted to the image array that is easy to compression after denoising the measured value that lacks redundancy and be difficult to compression, the new approaches of compressed sensing sampled measurement compression are provided; Make full use of the Given information of code aperture, associative transformation coding, saves compression bit; Utilize being closely connected of remainder matrix after transition coding and code aperture, introduce the Bit-Plane Encoding based on code aperture, further improved compression efficiency.

Accompanying drawing explanation

By reading the detailed description of non-limiting example being done with reference to the following drawings, it is more obvious that other features, objects and advantages of the present invention will become:

Fig. 1 is the compression frame flow chart based on mean filter conversion of the inventive method;

Fig. 2 is that the inventive method is based on code aperture bit-Plane Encoding schematic diagram.

Embodiment

Below in conjunction with specific embodiment, the present invention is described in detail.Following examples will contribute to those skilled in the art further to understand the present invention, but not limit in any form the present invention.It should be pointed out that to those skilled in the art, without departing from the inventive concept of the premise, can also make some distortion and improvement.These all belong to protection scope of the present invention.

Shown in Fig. 1-2, it is two matrixes by former sampled measurement matrix conversion that the present invention utilizes transition coding thought, then two matrixes is compressed separately to processing, by two megastages, it carried out to splitting and reorganizing and choose the algorithm adapting and compress, specifically:

First stage: conversion will be carried out as follows:

The discrete 3-D data set of step 1, L spectrum channel of definition is

n × M is the dimension of each spectrum channel image, and the dimension that makes Y represent that CASSI system obtains is the sampled measurement of N × (M+L-1), F _trepresent the image of selected reference spectra passage, random coded aperture matrix when R represents that CASSI system is used for producing Y;

Step 2, the element Y (i, j) measuring in matrix Y can be regarded as

f _l(x, y) represents the gray value of the pixel of l spectrum channel image, F _lrefer to respectively l the selecteed set of pixels of spectrum channel image and selecteed spectrum channel set with F.The addend number that forms each element in matrix Y can be obtained by code aperture matrix R:

$S=\left[\begin{array}{cccccccc}{R}_{11}& R_{11}+R_{12}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{\&Sigma;}}_{i=1}^L{R}_{1i}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{\&Sigma;}}_{i=j-L+1}^j{R}_{1i}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{\&Sigma;}}_{i=j-L+1}^M{R}_{1i}\\ R_{21}& R_{21}+R_{22}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{\&Sigma;}}_{i=1}^L{R}_{2i}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{\&Sigma;}}_{i=j-L+1}^j{R}_{2i}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{\&Sigma;}}_{i=j-L+1}^M{R}_{2i}\\ .& .& .& .& .& .& .& .\\ .& .& .& .& .& .& .& .\\ .& .& .& .& .& .& .& .\\ R_{N1}& {\mathrm{<figure-callout\; id="1"\; label="R\; N"\; filenames="HDA0000455176590000011.png"\; state="\{\{state\}\}">R</figure-callout>}}_N+R_{N2}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{\&Sigma;}}_{i=1}^L{R}_{\mathrm{Ni}}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{\&Sigma;}}_{i=j-L+1}^j{R}_{\mathrm{Ni}}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{\&Sigma;}}_{i=j-L+1}^M{R}_{\mathrm{Ni}}\end{array}\right]$

Step 3, according to mean filter thought, the measurement matrix after being converted is

it should be noted that the conversion obtaining like this

in each element in the scope of $ [0,255] $, or integer or decimal.Because the matrix that contains decimal can not be regarded image as, and directly compress decimal matrix and will cause information loss, the present invention will

be divided into two parts, i.e. business's part

with remainder part

Step 4, can prove

distribution and the distribution of original spectrum image very approaching, directly utilize Lossless Image Compression algorithm, as JPEG-LS and CALIC compress it.

Second stage: for remainder matrix

bit-Plane Encoding will carry out as follows:

Step 1, first will

in coefficient to be divided into 64 coefficients be that a rectangular block is processed.Each is divided into again three parts (section), and division principle is to make the coefficient of same section have similar positional structure and statistical property.

Step 2, use BitDepth_BloCk _mrepresent to describe the needed maximal bit figure place of amplitude of each coefficient y in m piece:

Step 3, in principle, all coefficients in m piece can be used BitDepth_Block _mindividual bit shows, but known according to the described transition coding based on code aperture of first stage, for the measured value after conversion

remainder part

the value of its capable j row of i must be in scope in, therefore this value has again ZeroPlane _ijbit on individual remarkable bit plane can correspondingly be defined as zero when given code aperture.Wherein:

Step 4, note bit plane b consist of b bit of all coefficient amplitude binary representations, and wherein, least significantly bit plane is designated as b=0.From the most remarkable bit plane b=BitDepth_Block _mto b=0, each bit plane of the m piece coefficient amplitude of encoding gradually.On each bit plane, by part, encode, it is one group that each part is once only got four positions, because some position is transformed encoding scheme and code aperture is constrained to zero, four bits are at most only contained in therefore each group, finally these bit groups are compressed with variable-length encoding.

Specific embodiment is below provided:

As shown in Figure 1, the transition coding process of the present embodiment comprises the steps:

Step 1, the dimension that CASSI system is obtained is that the sampled measurement Y of N × (M+L-1) is modeled as the pixel sum after the translation of different spectrum channels coding,

Step 2, according to code aperture matrix R, ask the addend number that forms each element in matrix Y:

$S=\left[\begin{array}{cccccccc}{R}_{11}& R_{11}+R_{12}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{\&Sigma;}}_{i=1}^L{R}_{1i}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{\&Sigma;}}_{i=j-L+1}^j{R}_{1i}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{\&Sigma;}}_{i=j-L+1}^M{R}_{1i}\\ R_{21}& R_{21}+R_{22}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{\&Sigma;}}_{i=1}^L{R}_{2i}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{\&Sigma;}}_{i=j-L+1}^j{R}_{2i}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{\&Sigma;}}_{i=j-L+1}^M{R}_{2i}\\ .& .& .& .& .& .& .& .\\ .& .& .& .& .& .& .& .\\ .& .& .& .& .& .& .& .\\ R_{N1}& {\mathrm{<figure-callout\; id="1"\; label="R\; N"\; filenames="HDA0000455176590000011.png"\; state="\{\{state\}\}">R</figure-callout>}}_N+R_{N2}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{\&Sigma;}}_{i=1}^L{R}_{\mathrm{Ni}}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{\&Sigma;}}_{i=j-L+1}^j{R}_{\mathrm{Ni}}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{\&Sigma;}}_{i=j-L+1}^M{R}_{\mathrm{Ni}}\end{array}\right]$

Step 3, according to mean filter thought, the measurement matrix after being converted is

be divided into business's part

with remainder part

two parts, right

directly utilize Lossless Image Compression algorithm (JPEG-LS or CALIC) to compress.

As shown in Figure 2, right in the present embodiment compression process

the Bit-Plane Encoding stage concrete implement to comprise following details:

Step 1, first will

in coefficient to be divided into 64 coefficients be a rectangular block (in figure, example is divided into four).Each has principle three parts (section) such as part 1, part 2, part 3 again of similar positional structure and statistical property according to coefficient;

Step 2, according to code aperture information R, determines

in the value of the capable j of i row

value on individual remarkable bit plane is all zero, therefore need not compress these bits.

Step 3, to m piece, from the most remarkable bit plane b=BitDepth_Block _mto each bit plane of gradual ground of b=0 code coefficient amplitude.On each bit plane, from part 1, to part 3, progressively encode, it is one group that each part is once only got four positions, because some position is defined as zero in step 2, therefore four bits are at most only contained in each group, finally these bit groups are compressed with variable-length encoding.

implementation result

According to above-mentioned steps, the present invention is that 24(is L=24 at a series of spectrum channels) the measured value of data set after via CASSI system on test, situation when respectively umber of beats (shots) being equaled to 4 and 6 has carried out analyzing experiment, transition coding based on mean filter and the Bit-Plane Encoding effect based on code aperture are assessed respectively, and compared with conventional efficient lossless compression method JPEG-LS and Calic.

The transition coding scheme based on mean filter that the present invention proposes is to have obtained the compression efficiency that is better than JPEG-LS approximately 4.6%, is better than Calic8 approximately 4.9% at 4 o'clock at umber of beats.Along with umber of beats increases, the advantage of transition coding is more obvious., introduce after the Bit-Plane Encoding based on code aperture, compression ratio has been got back and has been greater than 1% lifting meanwhile.Experiment shows, the method for utilizing code aperture information to carry out transition coding and Bit-Plane Encoding that the present invention proposes has effectively improved the Lossless Compression efficiency of CASSI systematic survey value.

Above specific embodiments of the invention are described.It will be appreciated that, the present invention is not limited to above-mentioned specific implementations, and those skilled in the art can make various distortion or modification within the scope of the claims, and this does not affect flesh and blood of the present invention.

Claims (5)

1. A compressed sensing spectral imaging measurement value compression method based on embedded transform coding, which is characterized in that the original sampling measurement value matrix is converted into two matrices by using the transformation coding idea, and then the two matrices are separately compressed. into two stages: In the first stage, the original sampling measurement value is converted into an image matrix that is more similar to the distribution of the original spectral channel image, and a random and sparse remainder matrix by using the mean filter transformation based on the coded aperture information; In the second stage, the positions where element bits are zero and bit planes are zero in the remainder matrix are determined by encoding the aperture information, and the remainder matrix is compressed efficiently by bit plane encoding. 2. The compressed sensing spectral imaging measurement value compression method based on embedded transform coding according to claim 1, characterized in that, in the first stage, fully utilize known Encoding aperture information, specifically: according to the random encoding aperture matrix used in the process of generating sampling measurement values by the system, calculate the number of addends for each element in the measurement value matrix, and use the transformation idea based on mean filtering to convert each element in the measurement value Elements are divided by the number of their addends to obtain an image matrix that is similar to the distribution of the original spectral channel image, and a random and sparse remainder matrix. 3. according to claim 1 or 2 described based on the compressed sensing spectrum imaging measurement value compression method of embedded transform coding, it is characterized in that, the image matrix that described first stage converts, the distribution of this matrix image and statistical characteristic and Similar to the original spectral channel image, it is directly compressed by the lossless image compression algorithm JPEG-LS and Calic. 4. The compressed sensing spectral imaging measurement value compression method based on embedded transform coding according to claim 1, characterized in that, specifically: in the second stage, determining element bits in the remainder matrix by encoding aperture information is zero and the position of the bit plane is zero, specifically: each element in the remainder matrix has some bits that can be determined to be zero according to the coded aperture information, and these bits that are determined to be zero will not be included in the compressed sequence. , directly recovered from the decoding end through the encoding aperture information; and the bits that are uncertain to be zero will be sequentially processed through the bit-plane encoding of bit-plane block and part. 5. The compressed sensing spectral imaging measurement value compression method based on embedded transform coding according to claim 1 or 4, wherein in the second stage, the highly efficient compression remainder matrix utilizing bit-plane coding is specifically : For the transformed remainder matrix, the bit-plane coding method based on coding aperture information is used, and the bit-plane coding and variable-length coding are used to compress the blocks according to the similar position structure and statistical characteristics, and the coding is performed block by block, starting from the most significant Bit plane to least significant bit plane progressively codes each bit plane of the coefficient magnitude, each bit plane is coded step by step from part 1 to part 3, and each part only takes four positions at a time as a group, because some positions According to the coding aperture information has been determined to be zero, so each subgroup contains only four bits at most, and finally these bit groups are compressed with variable length coding.

Images