CN102695055B - JPEG_LS (Joint Pho-tographic Experts Group-Lossless Standard) bit rate control method under high bit rate - Google Patents

Abstract

The invention discloses a JPEG_LS (Joint Pho-tographic Experts Group-Lossless Standard) bit rate control method under the high bit rate, which comprises the following implementing steps of: (1) dividing an original image into a plurality of strips; (2) processing all the strips by adopting a JPEG_LS lossless compression mode to obtain a bit stream subjected to lossless compression and a bit rate subjected to lossless compression; (3) judging whether to reach a total target bit rate; (4) carrying out processing in a bit rate control JPEG_LS lossy compression mode; and (5) outputting a final bit stream formed by combining the compressed bit stream with auxiliary information. According to the JPEG_LS (Joint Pho-tographic Experts Group-Lossless Standard) bit rate control method under high bit rate provided by the invention, the relation between the optimal quantization steps and the relation between target bit rates in the strips are searched by utilizing the rate-distortion characteristic of a residual error in each strip under the high bit rate in a JPEG_LS algorithm; by the utilization of the two relations, the time of searching the optimal quantization steps in a bit rate control algorithm is greatly simplified; and the JPEG_LS bit rate control method under the high bit rate is very suitable for high bit rate compression of a remote sensing image.

Description

JPEG_LS bit rate control method under high code check

Technical field

The present invention relates to technical field of image processing, further relate to JPEG_LS under a kind of high code check (JointPhotographic Experts Group LoSsless and near_lossless compression ofconituous_tone still image) bit rate control method, the present invention utilizes in JPEG_LS algorithm the rate distortion characteristic of residual error under high code check, find relation between image local zones of different target bit rate and the relation between optimum quantization step-length, then utilize lagrange's method of multipliers to find this region optimum quantization step-length by a local regional aim code check, relation based on quantization step quantizes it as overall quantization step.The present invention can be in the image compression encoding of various digital devices.

Background technology

JPEG_LS algorithm is the harmless and nearly Lossless Compression standard of rest image, because it has good performance under high code check, is widely used in Remote Sensing Image Compression.But there is a great defect in JPEG_LS algorithm, code check is uncontrollable.In remote sensing application, code check is controlled is a very important index, because no matter be that data leave on satellite or aircraft temporarily, or needs real-time down-transmitting data, and memory space and the design that passes down link bandwidth are all fixed.If code check changes very greatly, in order to prevent from overflowing, must design large memory space and data link bandwidth, such design can cause a large amount of wastes.If JPEG-LS algorithm has the controlled performance of code check, it will be the compression method that is very suitable for remote sensing images.

China Academy of Space Technology (Xi'an) its patent application " a kind of bit rate control method of JPEG-LS image compression " (number of patent application: 201010617932.2, publication number: disclose a kind of dynamic JPEG-LS bit rate control method CN102088602A).Xian Electronics Science and Technology University has proposed a kind of method of JPEG-LS being carried out to Rate Control similar with above method in its master thesis " research of JPEG-LS rate control algorithm ".These two kinds of methods exist an identical point to be, after picture portion, by the actual bit rate of coding region and the Accumulated deviation amount of target bit rate, utilize empirical value to need the quantization step of coding region after adjusting dynamically, thereby make the code check sum of All Ranges reach general objective code check.But the deficiency that these two kinds of methods still exist is, compression performance depends on empirical value, and it is more excellent that empirical value is got, and can obtain good compression performance, and it is poor that empirical value is got, and compression performance is also poor.And also there is another one shortcoming in these two kinds of methods, the mass discrepancy of the Recovery image of regional is larger, and the Recovery image quality in some region is better, and the Recovery image in some region is second-rate.

Summary of the invention

The object of the invention is to overcome the deficiency of above-mentioned prior art, lagrange's method of multipliers bit rate control method and JPEG_LS algorithm are combined, and make the method after combination there is good compression performance.

The thinking that realizes the object of the invention is, utilize in JPEG_LS algorithm the rate distortion characteristic of residual error under high code check, find respectively relation between image local zones of different target bit rate and the relation between optimum quantization step-length, then utilize lagrange's method of multipliers to find this region optimum quantization step-length by a local regional aim code check, the relation based between quantization step quantizes it as global optimum's quantization step.

For achieving the above object, method of the present invention comprises the steps:

(1) be divided into a plurality of bands

The original image of input is divided into N band, and the width of each band is identical with the width of original image, highly can get identical value and also can get different value, and N is more than or equal to 1 positive integer.；

(2) to all bands, adopt JPEG_LS Lossless Compression mode, obtain code stream after Lossless Compression and the code check of Lossless Compression;

(3) judge whether to reach general objective code check

Judge whether the code check after Lossless Compression reaches the general objective code check that user sets, when the code check of Lossless Compression is less than or equal to general objective code check, using the code stream after Lossless Compression as final compressed bit stream, select Lossless Compression mode as final compress mode, forward step (5) to; Otherwise, forward step (4) to;

(4) the JPEG_LS lossy compression method mode of Rate Control

4a) the prediction residual coefficient in extraction step (2);

4b) calculate the variance of the prediction residual coefficient of each band;

4c) by kurtosis value computing formula, obtain the kurtosis value of the prediction residual coefficient of first band;

4d) inquire about the relation table of kurtosis value and α, determine the value of parameter alpha in Generalized Gaussian Distribution Model;

4e) by following formula target bit rate computing formula, obtain the target bit rate of first band;

$r_1=r-\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\frac{{\mathrm{\β}}_i}{H}{\log }_{10}\left({\left(\frac{{\mathrm{\σ}}_1^2}{{\mathrm{\σ}}_i^2}\right)}^{m_R}\right)$

Wherein, r ₁be the target bit rate of first band, r is original image general objective code check, β _ithe percentage that the coefficient of i band accounts for original image coefficient, be the variance of the prediction residual coefficient of i band, H is the height of original image, and N is the band number after image that user chooses is cut apart, m _rbe the coefficient relevant to Generalized Gaussian Distribution Model, its value and α meet one-to-one relationship, and ∑ is that the company in mathematics puts in marks;

4f) obtain the optimum quantization step-length of first band: the target bit rate of first band is controlled lagrange's method of multipliers, obtain the optimum quantization step-length of first band;

4g) obtain the optimum quantization step-length of each band: under high code check, the optimum quantization step-length using the optimum quantization step-length of first band as each band;

4h) lossy compression method: use the optimum quantization step-length of each band, each band is carried out to JPEG_LS lossy compression method;

(5) the final code stream that output squeezing code stream and supplementary are combined.

Compared with prior art, tool has the following advantages in the present invention:

First, because the present invention has carried out detailed analysis to the rate distortion characteristic of prediction residual in JPEG_LS algorithm, found for the distinctive rate distortion characteristic of JPEG_LS algorithm under high code check, overcome the mode of utilizing empirical value in prior art, made the present invention there is good rate control accuracy and good compression performance.

Second, because the present invention has found the relation between different band optimum quantization step-lengths, all bands have identical optimum quantization step-length, overcome the optimum quantization step-length that needs to calculate respectively each band in prior art, make the present invention greatly save the time of finding optimum quantization step-length in Rate Control technology, only need to find its optimum quantization step-length to a band, need not all find this complex operations of optimum quantization step-length to all bands

The 3rd, because the present invention has found the relation between different band target bit rates, by general objective code check, obtained the target bit rate of first band, only first band is carried out to Lagrangian Rate Control, overcome in prior art and needed each band to carry out Lagrangian Rate Control, made the present invention when greatly simplifying Rate Control, guarantee the precision of Rate Control.

The 4th, because considering remote sensing images, the present invention there is the simple especially image of some extra high Compression requirements and some atural objects, first adopt JPEG_LS Lossless Compression, while only having code check when Lossless Compression higher than general objective code check, just adopt the controlled lossy compression method of code check.In the case, the introducing of Lossless Compression has improved compression performance greatly.

Accompanying drawing explanation

Fig. 1 is flow chart of the present invention;

Fig. 2 is the resolution chart of emulation experiment of the present invention.

Embodiment

Below in conjunction with 1 pair of performing step of the present invention of accompanying drawing, be described in detail.

Step 1. is divided into N band by the original image of input, and the width of each band is identical with the width of original image, highly can get identical value and also can get different value, and N is more than or equal to 1 positive integer.

Step 2. pair all bands, adopt JPEG_LS Lossless Compression mode, obtain code stream after Lossless Compression and the code check of Lossless Compression.JPEG_LS Lossless Compression mode is that use standard JPEG_LS algorithm compresses image, and after prediction, residual error coefficient does not quantize.

Step 3. judges whether the code check after Lossless Compression reaches the general objective code check that user sets, and general objective code check is that user is according to the requirements set of practical application fixed value.When the code check of Lossless Compression is less than or equal to general objective code check, using the code stream after Lossless Compression as final compressed bit stream, select Lossless Compression mode as final compress mode, forward step 5 to; Otherwise, forward step 4 to.

The JPEG_LS lossy compression method mode of step 4. Rate Control

Rate Control namely bit rate is distributed, and it is very important Yi Ge branch in coding theory, is also the study hotspot of coding theory.Its basic point of departure is, the zones of different of piece image is distributed to different code checks, in the situation that total bitrate is certain, makes the distortion of entire image minimum.Can adopt following formula to explain:

$\min \left\{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{D}_i\left(R_i\right)\right\}$

$\mathrm{subject\; to}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{R}_i=R$

Wherein, R is given general objective bit number, R _ithe target bit of i region allocation, D _i(R _i) be the distortion in i region, N is total number in region.This is a constrained optimization problem, is very difficult to solve.

Lagrange's method of multipliers is converted to a unconfined optimization problem by this problem, as following formula:

$\min \left\{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}(D_i\left(R_i\right)+\mathrm{\λ}{R}_i)\right\}$

Wherein, λ is the Lagrangian coefficient in lagrange's method of multipliers, and R is given general objective bit number, R _ithe target bit of i region allocation, D _i(R _i) be the distortion in i region, N is total number in region.

For unconfined optimization problem, a given λ (>=0) value must find the R of one group of optimum _i(i=1 ... N), if meet general objective bit number, this N R now _iit is exactly the solution of above formula.If be discontented with Football Association's target bit, upgrade the value of λ, continue to select another to organize optimum R _i(i=1 ... N), until meet general objective bit number.The key of whole lagrange's method of multipliers need to find the relation of distortion and bit number exactly, and value how to upgrade λ.

The lagrange's method of multipliers of take in the present invention carries out Rate Control as basis to JPEG_LS algorithm.Because the relation of distortion and code check is the key of Rate Control, what in JPEG_LS, need to quantize is the residual error coefficient after prediction, so first analyzed the rate distortion function of residual error coefficient in JPEG_LS.

Many documents have all been analyzed prediction residual after normalization and have been obeyed generalized Gaussian distribution (GGD), as following formula:

$p\left(x\right)=\frac{\mathrm{\α}}{2\mathrm{\σ\Γ}(1/\mathrm{\α})}\sqrt{\frac{\mathrm{\Γ}(3/\mathrm{\α})}{\mathrm{\Γ}(1/\mathrm{\α})}}\exp \{-{\left(\sqrt{\frac{\mathrm{\Γ}(3/\mathrm{\α})}{\mathrm{\Γ}(1/\mathrm{\α})}}\left(\frac{\left|x\right|}{\mathrm{\σ}}\right)\right)}^{\mathrm{\α}}\}$

Wherein, p (x) is the probability density function of prediction residual after normalization, and α is exponential decay rate, and σ is the standard deviation of prediction residual coefficient, and Γ () is gamma function.And many documents provide coefficient after wavelet transformation generally all can meet a kind of in distributing of five kinds of GGD, i.e. α={ 0.5,0.75,1.0,1.5,2.0}.In reality, can first calculate the kurtosis value K of coefficient, then utilize following table " relation table of kurtosis value and α ", table look-up and obtain the value of α.

The span of kurtosis value K	α
		-∞＜＝K＜3.3010	0.5
3.3010＜＝K＜7.8250	0.75
		3.8250＜＝K＜14.8810	1.0
14.8810＜＝K＜17.4250	1.5
		17.4250＜＝K＜+∞	2.0

In fact the high frequency coefficient of small echo is similar to the residual error coefficient after prediction, so generally the residual error coefficient after prediction also meets a kind of in distributing of five kinds of GGD.Multirow continuous in JPEG_LS algorithm is as a band, and each band is a region, and each band also meets GGD and distributes.By above analysis, can be seen, now to the Rate Control of JPEG_LS algorithm, just can be converted to following lagrange's method of multipliers:

$\min \{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\β}}_i{\mathrm{\σ}}_i^2{d}_i\left(r_i\right)+\mathrm{\λ}{\mathrm{\β}}_i{r}_i)\}$

β _ir _iWH＝R _i，β _id _i(r _i)WH＝D _i(R _i)

Wherein, λ is the Lagrangian coefficient in lagrange's method of multipliers, β _ithe percentage that the coefficient of i band accounts for original image coefficient, R _ithe target bit that i band distributes, r _ithe code check that each pixel of i band is distributed, d _i(r _i) be the distortion of corresponding each pixel, be the variance of the prediction residual coefficient of i band, W is the width of original image, and H is the height of original image, and N is the band number after image that user chooses is cut apart.

By solving Lagrange's multiplier problem, i the code check r that each pixel of band is distributed _ibe function.Same, the quantization step Δ after residual error coefficient normalization _ialso be function.Because residual error coefficient meets GGD model, under high code check, obtain the relational expression of following two simplification:

$r_i=m_R\log (\mathrm{\λ}/{\mathrm{\σ}}_i^2)+b_R$

$\log \left({\mathrm{\Δ}}_i\right)=m_{\mathrm{\Δ}}\log (\mathrm{\λ}/{\mathrm{\σ}}_i^2)+b_{\mathrm{\Δ}}$

Wherein, λ is the Lagrangian coefficient in lagrange's method of multipliers, r _ithe bit rate that each pixel of i band is distributed, Δ _ithe quantization step obtaining after residual error coefficient normalization, the variance of the prediction residual coefficient of i band, m _r, b _r, m _Δ, b _Δall depend on the parameter alpha in GGD model, its value and α meet one-to-one relationship.Conventionally a look-up table setting up corresponding relation between these values and α, is worth accordingly by tabling look-up.When the α of different bands is identical, while namely obeying univesral distribution, these variablees of all bands are all identical.Δ _ibe the quantization step obtaining after residual error coefficient normalization, the quantization step of actual each band is

Under a kind of high code check of the present invention, JPEG_LS bit rate control method is exactly according to the feature of JPEG_LS algorithm itself, and Lagrangian rate control algolithm is applied to JPEG_LS, realizes good Rate Control effect, and the detailed step of Rate Control is as follows,

The first step, the prediction residual coefficient in extraction step 2;

Second step, calculates the variance of the prediction residual coefficient of each band

The 3rd step, by kurtosis value computing formula, obtains the kurtosis value of the prediction residual coefficient of first band, and kurtosis value is calculated by following formula,

$K=\frac{\underset{j=1}{\overset{M}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x_j})}^4}{{\mathrm{\σ}}_1^4}$

Wherein, K is the kurtosis value of first band, x _jthe prediction residual coefficient of first band, the mean value of the prediction residual coefficient of first band, be the biquadratic of the prediction residual coefficient standard deviation of first band, M is the number of the prediction residual coefficient of first band;

The 4th step, obtains after kurtosis value K, by searching " relation table of kurtosis value and α ", is obtained the value of GGD Model Parameter α by the interval at kurtosis value K place;

The 5th step, by following formula target bit rate computing formula, obtains the target bit rate of first band;

$r_1=r-\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\frac{{\mathrm{\β}}_i}{H}{\log }_{10}\left({\left(\frac{{\mathrm{\σ}}_1^2}{{\mathrm{\σ}}_i^2}\right)}^{m_R}\right)$

Wherein, r ₁be the target bit rate of first band, r is original image general objective code check, β _ithe percentage that the coefficient of i band accounts for original image coefficient, be the variance of the prediction residual coefficient of i band, H is the height of original image, and N is the band number after image that user chooses is cut apart, m _rbe the coefficient relevant to Generalized Gaussian Distribution Model, its value and α meet one-to-one relationship, and ∑ is that the company in mathematics puts in marks;

The 6th step, the target bit rate of first band is controlled lagrange's method of multipliers, obtains the optimum quantization step-length of first band;

The 7th step, under high code check, using first band optimum quantization step-length as each band optimum quantization step-length;

Entire image is divided into after N band, adopts the optimum quantization step-length of each band obtaining after lagrange's method of multipliers to be (i=1,2 ..., N), the optimum quantization step-length of each band is analyzed, meet following formula:

${\mathrm{\Δ}}_i^{*}={\mathrm{\σ}}_i{(\mathrm{\λ}/{\mathrm{\σ}}_i^2)}^{m_{\mathrm{\Δ}}}*{10}^{b_{\mathrm{\Δ}}}$

Wherein, λ is the Lagrangian coefficient in lagrange's method of multipliers, the standard deviation of the prediction residual coefficient of i band, m _Δ, b _Δall depend on the parameter alpha in GGD model, its value and α meet one-to-one relationship.For five kinds of GGD distributed models conventional in table " relation table of kurtosis value and α ", m _Δidentical and m _Δ=1/2.After abbreviation, above formula becomes following form:

${\mathrm{\Δ}}_i^{*}=\sqrt{\mathrm{\λ}}*{10}^{b_{\mathrm{\Δ}}}$

Wherein, be the optimum quantization step-length of i band, λ is the Lagrangian coefficient in lagrange's method of multipliers, b _Δbe the coefficient relevant to Generalized Gaussian Distribution Model, its value and α meet one-to-one relationship;

By a large amount of test, find to adopt the residual error coefficient after JPEG_LS coding to be divided into after a plurality of bands with piece image, during the GGD of each band distributes, the value of α is almost identical, so in the present invention, supposes that each band all obeys unified GGD model.Because the GGD distribution of each band is identical, the optimum quantization step-length obtaining is also identical, so the present invention adopts the optimum quantization step-length of first band as the optimum quantization step-length of each band;

The 8th step, is used the optimum quantization step-length of each band, and each band is carried out to JPEG_LS lossy compression method.

The final code stream that step 5. output squeezing code stream and supplementary are combined, compressed bit stream refers to, when adopting Lossless Compression mode, is Lossless Compression code stream, when adopting lossy compression method mode, is lossy compression method code stream.Supplementary refers to, comprises compress mode information and optimum quantization step-length information

Effect of the present invention can be described further by following emulation experiment.

Software adopts Microsoft Visual C++6.0 Integrated Development software and C language to realize on the Windows7 of Microsoft company environment.The present invention has chosen 3 width and has taken in the remote sensing images of Beijing Area as shown in Figure 2, and the size of 3 width images is respectively 1024 * 1024, adopts 8 bit storage, and Fig. 2 (a) is for taking the photograph the remote sensing figure in mountain region, forest, river; Fig. 2 (b) is for taking the photograph the remote sensing figure in city; Fig. 2 (c) is for taking the photograph the remote sensing figure in suburb, city.

The performance of put forward the methods in order to assess, has selected three kinds of algorithms, is respectively method of the present invention, JPEG2000 canonical algorithm KDU, prior art " the rate control algorithm JPEG_LS_Q in Xian Electronics Science and Technology University's master thesis ".The method that the present invention proposes adopts two equal bands.Having tested respectively three kinds of methods is code check (bpp) and Recovery image Y-PSNR (PSNR) value of 2,3,4 o'clock at compression multiple.For different compression ratios, experimental result is presented in following table respectively.

From upper table, can see, three kinds of methods can reach good Rate Control, and when code check is very high, during as 2 times of compressions, the method that the present invention proposes can realize Lossless Compression, and other two kinds of algorithms only have parts of images to realize Lossless Compression.Owing to considering, in remote sensing images, there is the simple especially image of some extra high Compression requirements and some atural objects, first the method that the present invention proposes considers to adopt Lossless Compression, while only having code check when Lossless Compression higher than general objective code check, just consider to adopt the controlled lossy compression method of code check.Relatively PSNR finds, the compression performance of JPEG-LS under high code check is better than JPEG2000, and the method that the present invention proposes has best mean P SNR.Even JPEG_LS_Q compares with prior art, code check is similar, and the method that the present invention proposes still has higher PSNR.This is because the method that the present invention proposes is to obtain the target bit rate of first band according to the statistical property of each band, and then obtains optimum quantization step, and all bands have adopted identical quantization step.And prior art JPEG_LS_Q need to ceaselessly change the quantization step of different bands, control code check, this will cause the recovery effects of some band fine, and the recovery effects of some band is very poor, has reduced the performance of integral image.

While more than testing, the band number in the method that the present invention proposes is chosen as 2, in reality, can adopt more bands.Along with the increase of band number, the precision of Rate Control can be worse and worse, and the time that still coding needs can be shorter and shorter, so can select according to the actual requirements the number of band.

The present invention proposes a kind of method of JPEG_LS being carried out Rate Control under high code check.The method has been utilized the similitude of residual distribution in each band, the relation between optimum quantization step-length and the relation between target bit rate in each band have been found, utilize this two groups of relations, greatly simplify the time of finding optimum quantization step-length in rate control algorithm, be very suitable for the compression of remote sensing images high bit rate.

Claims (3)

1. a JPEG_LS bit rate control method under high code check, comprises the steps:

(1) original image of input is divided into N band, the width of each band is identical with the width of original image, highly can get identical value and also can get different value, and N is more than or equal to 1 positive integer;

(2) to all bands, adopt JPEG_LS Lossless Compression mode, obtain code stream after Lossless Compression and the code check of Lossless Compression, and the residual error coefficient after being predicted; Described JPEG_LS Lossless Compression mode is that use standard JPEG_LS algorithm compresses image, and after prediction, residual error coefficient does not quantize;

(3) judge that whether the code check after Lossless Compression reaches the general objective code check that user sets, and when the code check of Lossless Compression is less than or equal to general objective code check, using the code stream after Lossless Compression as final compressed bit stream, forwards step (5) to; Otherwise, forward step (4) to;

(4) the JPEG_LS lossy compression method mode of Rate Control;

4a) the prediction residual coefficient in extraction step (2);

4b) calculate the variance of the prediction residual coefficient of each band;

4c) by kurtosis value computing formula, obtain the kurtosis value of the prediction residual coefficient of first band;

Described kurtosis value computing formula is as follows,

$K=\frac{\underset{j=1}{\overset{M}{\mathrm{\Σ}}}{(x_j-\stackrel{\‾}{x_j})}^4}{{\mathrm{\σ}}_1^4}$

Wherein, K is the kurtosis value of first band, x _jthe prediction residual coefficient of first band, the mean value of the prediction residual coefficient of first band, be the biquadratic of the prediction residual coefficient standard deviation of first band, M is the number of the prediction residual coefficient of first band;

4d) inquire about the relation table of kurtosis value and α, determine the value of parameter alpha in Generalized Gaussian Distribution Model;

4e) by following target bit rate computing formula, obtain the target bit rate of first band;

$r_1=r-\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\frac{{\mathrm{\β}}_i}{H}{\log }_{10}\left({\left(\frac{{\mathrm{\σ}}_1^2}{{\mathrm{\σ}}_i^2}\right)}^{m_R}\right)$

Wherein, r ₁be the target bit rate of first band, r is original image general objective code check, β _ithe percentage that the coefficient of i band accounts for original image coefficient, be the variance of the prediction residual coefficient of i band, H is the height of original image, and N is the band number after image that user chooses is cut apart, m _rbe the coefficient relevant to Generalized Gaussian Distribution Model, its value and α meet one-to-one relationship, and ∑ is that the company in mathematics puts in marks;

4f) obtain the optimum quantization step-length of first band, concrete steps are as follows:

The first step, value λ>=0 of a given Lagrangian coefficient lambda, finds the R of one group of optimum _i(i=1 ... N), if λ is optimum λ value, otherwise continues to upgrade the value of λ, until meet

Second step, is tabled look-up and is obtained b by the value of α _Δvalue;

The 3rd step, utilizes formula obtain optimum quantization step;

Wherein, λ is the Lagrangian coefficient in lagrange's method of multipliers, and R is given general objective bit number, R _ibe the target bit of i region allocation, α is generalized Gaussian distribution exponential decay rate, b _Δdepend on the parameter alpha in generalized Gaussian distribution GGD model, its value and α meet one-to-one relationship;

4g) under high code check, the optimum quantization step-length using the optimum quantization step-length of first band as each band;

4h) use the optimum quantization step-length of each band, each band is carried out to JPEG_LS lossy compression method, described JPEG LS lossy compression method mode is that use standard JPEG LS algorithm compresses image, and after prediction, residual error coefficient quantizes;

(5) the final code stream that output squeezing code stream and supplementary are combined; Described supplementary comprises compress mode information and optimum quantization step-length information.

2. JPEG_LS bit rate control method under high code check according to claim 1, it is characterized in that, step 4d) relation table of kurtosis value described in and α is the look-up table of the corresponding relation of parameter alpha in the reflection span of kurtosis value and Generalized Gaussian Distribution Model.

3. JPEG_LS bit rate control method under high code check according to claim 1, it is characterized in that, compressed bit stream described in step (5) refers to, when adopting Lossless Compression mode, is Lossless Compression code stream, when adopting lossy compression method mode, is lossy compression method code stream.