CN101754021B - Method for realizing mobile phone mobile portal technology based on improved wavelet-transform image compression method - Google Patents

Abstract

The invention relates to mobile internet technical field, in particular to a mobile phone mobile portal method based on improved wavelet-transform image compression method. The portal technology includes an encoding part and a decoding part; the concrete procedures are as follows: the encoding part is composed of three basic modules, namely a wavelet transform, quantification and entropy encoding, wherein the original image is divided into blocks, each image block improves the wavelet transform independently and performs relative quantification and entropy encoding, and then the encoding streams of all image blocks are written into bit streams one by one according to ordered scanning sequence; in decoding end, the data of each image block in the bit stream is performed with entropy decoding, and is written into the corresponding positions of the whole image one by one according to the position relationship of the wavelet coefficient of each image block and the whole image wavelet coefficient, and then the image is recovered through invert quantification and invert improved wavelet transform. The system has the characteristics of easy software and hardware implementation, and pixel scalability.

Description

Based on improving the method that wavelet-transform image compression method realizes mobile phone mobile portal

Technical field

The present invention relates to mobile internet technical field, especially a kind of based on improving the method that wavelet-transform image compression method realizes mobile phone mobile portal.

Background technology

Mobile Internet has become up-to-date hot technology, along with popularizing and the propelling of 3G network of smart mobile phone, has increasing application and uses thereon.But because the restriction of cell phone platform, its function and disposal ability can limit to some extent, can not so many software be installed as computer, such as: on mobile phone, realize the reading of pdf, need the software support of Adobe company, realize the reading of word, the support that then needs to install the Office of Microsoft; Even if support the mobile phone of windows Mobile operating system that Office word and IE have been installed, cannot be mentioned in the same breath but compare still on its function and the computer.And because the popularizing also for some time of 3G network and mobile phone, though the network of at present a large amount of 2G is very ripe, network speed is limited.Present mobile phone door adopts the form of wap or http agreement mostly, browses.At the information portal, this pattern is recommendable, because its function is mainly browsed.But for enterprise portal, especially need the door of the application (as OA system, PDM system) of integrated complicated applications system (as some complicated ERP system) and a large amount of annexes, adopt the form of wap form and http agreement, just can not realize based on C/S.And utilize improved wavelet image compress technique, and realize control to computer, the screen content of computer by the split screen compress mode, is presented on the mobile phone, thereby can realizes mobile phone mobile portal.

The notion of wavelet transformation is at first to be proposed in 1974 by the engineer J.Morlet that France is engaged in the oil signal processing.It is compared with Fourier conversion, window Fourier conversion (Gabor conversion), this is the local conversion of time and frequency, thereby can effectively information extraction from signal, by calculation functions such as flexible and translations function or signal are carried out multiscale analysis (Multiscale Analysis), the indeterminable many difficult problems of Fourier conversion have been solved, the small echo variation is described as " school microscop ", and it is a landmark progress on the harmonic analysis development history.

Wavelet transformation theory is time (sky) frequency-domain analysis theory of rising in recent years, by original image is carried out wavelet transformation, picture signal can be transformed to wavelet field by time-domain (spatial domain) expression and represent.Utilize the quadrature or the biorthogonal conversion characteristic of wavelet transformation, remove the correlation between image pixel, the removal of images signal is in the redundancy in space, and the energy of concentrated picture signal, for the coefficient quantization of back, coefficient bits modeling, arithmetic coding etc. provide prerequisite, for image encoding efficiently lays the foundation.Traditional convolution small echo (first generation small echo) conversion, owing to adopt the convolution algorithm method, the process complexity, operand is big, and real-time is relatively poor, is unfavorable for the realization of hardware.Nineteen ninety-five Sweldens has proposed a kind of new wavelet construction method that does not rely on Fourier transform---and Via Lifting Scheme (Lifting scheme) is referred to as second generation wavelet transformation.This Via Lifting Scheme has not only kept the characteristic of first generation small echo, has overcome its translation and flexible consistency simultaneously again.Many Chinese and overseas scholars have carried out broad research to the Via Lifting Scheme wavelet transformation, have obtained great successes.

The small echo lifting scheme provides a kind of new implementation method faster for first generation wavelet transformation.Method for improving provides an effectively method of structure nonlinear wavelet, and the nonlinear wavelet that constructs is compared with traditional wavelet transformation, calculates simply fast, and is suitable for the conversion to integer of self adaptation, non-linear, nonsingular sampling and integer.Bi-orthogonal wavelet transformation is widely used in the image compression field because having linear property, studies have shown that at present the bi-orthogonal wavelet transformation of any FIR of having structure can be obtained by lifting and the antithesis lifting process that the inertia conversion replaced through the limited step.

The complexity of small echo lifting scheme has only about half of original convolution method, and real-time is good, and computing is simple, therefore becomes the main stream approach of calculating wavelet transform.It is the focus of studying recently that wavelet transform promotes the fast algorithm of realizing.Popular wavelet coefficient coding method at present has EZW algorithm, spiht algorithm and the EBCOT algorithm that adopted by JPEG2000 etc.These algorithms are the EBCOT algorithm particularly, all has superior performance, adopts in a large number.But they still have weak point, show:

1, these algorithms, especially EBCOT can realize the effective compression to image, and produced simultaneously code stream has resolution flexible (resolution scalability), signal to noise ratio scalability (SNRscalability) and arbitrary access outstanding characteristics such as (random access).But in this algorithm, the selection of code block all is with carrying out in the subband identical in the one-level frequency band in wavelet field.Though the mode of choosing like this can realize the characteristic of resolution flexible, fail to make full use of between the coefficient of all directions subband under the same resolution and the correlation between the coefficient of same direction different resolution.

2, existing algorithm all is the Static Compression technology, and the image that shows on the mobile phone is the part of whole screen, and user's operation also is the part of screen, and present algorithm can't compress the dynamic area separately.

Summary of the invention

Can't compress the deficiency of dynamic area separately in order to overcome existing wavelet-transform image compression method, the invention provides a kind of based on improving the method that wavelet-transform image compression method realizes mobile phone mobile portal.

The technical solution adopted for the present invention to solve the technical problems is: a kind of method based on improvement wavelet-transform image compression method realization mobile phone mobile portal, comprise coded portion and decoded portion, and concrete steps are as follows:

Coded portion is made up of wavelet transformation, quantification and three basic modules of entropy coding;

Described wavelet transformation refers to carry out earlier image segmentation, and original image is carried out piecemeal, carries out the coefficient reorganization again, the sub-band coefficients of each image block, according to the identical principle of location index, add to one by one in the respective sub-bands of entire image, then to the independent lifting wavelet transform of each image block;

Described quantification refers to the floating data after the lifting wavelet transform is carried out quantification treatment, so that wavelet coefficient is mapped to integer field from the floating-point territory;

Described entropy coding refers to that by improved spiht algorithm and arithmetic coding quantized transform coefficients being changed into one is used for transmitting or the compressed bit stream of storing, and writes bit stream to the encoding stream of all images piece one by one according to orderly scanning sequency then;

Described improved spiht algorithm adopts the tree structure of EZW algorithm, and all wavelet tree chunks are carried out level scanning, has scanned all tree root pieces earlier, scans the next level of all trees again; The perhaps first coefficient block in the scanning lowest frequency subband, the order that adopts raster scan again is to scanning in the same subband of one deck; Perhaps scan the coefficient block in the high-frequency sub-band earlier, the order that adopts raster scan again is to scanning in the same subband of one deck;

In decoding end, after the data of each image block in the bit stream are carried out entropy decoding, concern, write the relevant position of entire image one by one according to the wavelet coefficient and the position between the entire image wavelet coefficient of image block, carry out re-quantization and contrary lifting wavelet transform then, just can recover image.

According to another embodiment of the invention, further comprise each index block in the described original image through behind wavelet transformation, its each subband index piece lays respectively on the identical index position of respective sub-bands behind the whole original image wavelet transformation.

According to another embodiment of the invention, number of transitions n 〉=3 that further comprise described lifting wavelet transform.

According to another embodiment of the invention, comprise that further described quantification is that the floating data after the lifting wavelet transform is carried out quantification treatment, so that wavelet coefficient is mapped to integer field from the floating-point territory;

If O (r) expression visual direction quantizing factor, expression formula is

$O\left(r\right)=2^{R_r},$ r＝0，1，2，3

R wherein _rThe index of representing the visual direction quantizing factor of different directions on the same level of resolution, and

R ₀+R ₁+R ₂+R ₃＝32

They take the space of 4 bytes altogether, when storage and transmission, flow to decoder prior to code stream.

S (k) expression vision band quantizing factor, expression formula is

S(k)＝2 ^k-1，k＝0，1，2，3

Wherein r represents the frequency band of different directions on the same level of resolution, and k represents level of resolution; We can set the quantization step Δ of the subband on the r direction on the k class resolution ratio grade thus _{K, r}For

${\mathrm{\Δ}}_{k,r}=\mathrm{\α}\×O\left(r\right)\×S\left(k\right)=\mathrm{\α}\×2^{R_r+k-1}$

Wherein α represents to quantize to adjust the factor; Quantize to adjust the factor and can carry out adaptive adjustment, for example can be set as follows according to the visual importance of current block b according to the characteristic of current coefficient block b:

$\mathrm{\α}=1+\underset{k=0}{\overset{4^{N-M}-1}{\mathrm{\Σ}}}{\mathrm{TB}}_k/(4^{N-M}\×{\mathrm{TB}}_b)$

In addition, for the sake of simplicity, can also directly to make α be a constant or directly get α=1.

Then quantization operation is with the wavelet coefficient y of subband b _b(u v) is quantified as quantization parameter q _b(u, v):

The re-quantization process is,

y _b(u，v)＝q _b(u，v)×Δ _k，r。

The invention has the beneficial effects as follows,

1, each image block carries out transition coding separately, thereby can realize parallel processing and coding between image block.

2, adopt the algorithm of image segmentation conversion, the demand to internal memory during computing is less.The continuous multiplexing lifting wavelet transform that just can finish entire image of little internal memory.

3, image block reconfigures the wavelet coefficient that obtains entire image according to the coefficient corresponding relation then through behind the wavelet transformation.Do like this and can reduce the bigger internal memory of frequent use in the conversion process.Improve the internal memory utilization ratio.

4, because improved spiht algorithm also can form the embedding bit stream, when bit stream interrupts in the arbitrfary point, can reconstructed image.Therefore control bit speed and compression ratio easily.

5,, can realize ROI zone specific coding because improved spiht algorithm carries out the subband block encoding.

6, improved spiht algorithm and lifting wavelet transform algorithm all are easy to realize with software and hardware.Therefore this system also possesses the characteristic that software and hardware is easy to realize.

7, because encryption algorithm is to encode according to level of resolution, code stream is organized according to the mode that resolution increases progressively, so this system has resolution flexible.

What 8, improved spiht algorithm generated still is embedded bitstream, thereby can realize progressive decoding and progressive demonstration.

9, improved spiht algorithm is compared with traditional spiht algorithm, has higher compression ratio under the identical PSNR, can improve 10%-20% through evidence

Description of drawings

The present invention is further described below in conjunction with drawings and Examples.

Fig. 1 is the general flow figure of Wavelet image encoding and decoding;

Fig. 2 is that the small echo volume is separated system block diagram;

Fig. 3 is that size is 2 * 2 wavelet tree block structures;

Among Fig. 4, (a) be spiht algorithm and improved spiht algorithm subband and wavelet coefficient scanning sequency; (b) be the scanning sequency of spiht algorithm in the identical frequency band; (c) be the scanning sequency of improved spiht algorithm in the identical frequency band.

Embodiment

The present invention is further detailed explanation with preferred embodiment in conjunction with the accompanying drawings now.These accompanying drawings are the schematic diagram of simplification, basic structure of the present invention only is described in a schematic way, so it only show the formation relevant with the present invention.

As Fig. 1 is structural representation of the present invention, and a kind of method based on improvement wavelet-transform image compression method realization mobile phone mobile portal comprises coded portion and decoded portion, and concrete steps are as follows:

Coded portion mainly is made up of wavelet transformation, quantification and three basic modules of entropy coding, earlier original image is carried out piecemeal, the independent lifting wavelet transform of each image block, quantification of being correlated with again and entropy coding write bit stream to the encoding stream of all images piece one by one according to orderly scanning sequency then;

In decoding end, after the data of each image block in the bit stream are carried out entropy decoding, concern, write the relevant position of entire image one by one according to the wavelet coefficient and the position between the entire image wavelet coefficient of image block, carry out re-quantization and contrary lifting wavelet transform then, just can recover image.

Each index block in the described original image is through behind wavelet transformation, and its each subband index piece lays respectively on the identical index position of respective sub-bands behind the whole original image wavelet transformation.

The number of transitions n of described lifting wavelet transform 〉=3.

Described quantification is that the floating data after the lifting wavelet transform is carried out quantification treatment, so that wavelet coefficient is mapped to integer field from the floating-point territory;

If O (r) expression visual direction quantizing factor, expression formula is

$O\left(r\right)=2^{R_r},$ r＝0，1，2，3

R wherein _rThe index of representing the visual direction quantizing factor of different directions on the same level of resolution, and

R ₀+R ₁+R ₂+R ₃＝32

They take the space of 4 bytes altogether, when storage and transmission, flow to decoder prior to code stream.

S (k) expression vision band quantizing factor, expression formula is

S(k)＝2 ^k-1，k＝0，1，2，3

Wherein r represents the frequency band of different directions on the same level of resolution, and k represents level of resolution; We can set the quantization step Δ of the subband on the r direction on the k class resolution ratio grade thus _{K, r}For

${\mathrm{\Δ}}_{k,r}=\mathrm{\α}\×O\left(r\right)\×S\left(k\right)=\mathrm{\α}\×2^{R_r+k-1}$

Wherein α represents to quantize to adjust the factor; Quantize to adjust the factor and can carry out adaptive adjustment, for example can be set as follows according to the visual importance of current block b according to the characteristic of current coefficient block b:

$\mathrm{\α}=1+\underset{k=0}{\overset{4^{N-M}-1}{\mathrm{\Σ}}}{\mathrm{TB}}_k/(4^{N-M}\×{\mathrm{TB}}_b)$

In addition, for the sake of simplicity, can also directly to make α be a constant or directly get α=1.

Then quantization operation is with the wavelet coefficient y of subband b _b(u v) is quantified as quantization parameter q _b(u, v):

The re-quantization process is,

y _b(u，v)＝q _b(u，v)×Δ _k，r。

Described entropy coding is by improved spiht algorithm and arithmetic coding quantized transform coefficients to be changed into one to be used for transmitting or the compressed bit stream of storing.

The general flow of Wavelet image encoding and decoding as shown in Figure 1.Coded portion mainly is made up of wavelet transformation (wavelet transform), quantification (quantization) and entropy coding three basic modules such as (entroy encoding).Decoded portion is then carried out along opposite process.

Each index block in the original image is through behind wavelet transformation, and its each subband index piece lays respectively on the identical index position of respective sub-bands behind the whole original image wavelet transformation.Utilize this conclusion, we can carry out piecemeal to image earlier, and each image block carries out lifting wavelet transform and relevant quantization encoding operation separately.According to certain scanning sequency the encoding stream of all images piece is write bit stream one by one then.In decoding end, after the data decode of each image block in the code stream,, write the relevant position of entire image one by one according to the wavelet coefficient and the relation of the position between the entire image wavelet coefficient of image block, carry out re-quantization and contrary lifting wavelet transform then, just can recover original image.The whole system block diagram as shown in Figure 2.

Fig. 2 provides is processing to the single image component, and for the image of a plurality of components, the encoding process of each component is separate, so this figure is equally applicable to multicomponent situation.Adopt 3 grades of lifting wavelet transform in the native system.

1. image segmentation and coefficient reorganization

Because after through n (generally getting n 〉=3) level lifting wavelet transform, each 1 * 1 coefficient block in the lowest frequency subband all corresponding in the original image size on the same spatial location be 2 ⁿ* 2 ⁿImage block.Therefore, if after the image segmentation, the size 2 of each image block ^M* 2 ^M, then 2 ^MMinimum should get 8, and maximum rounds the height or the width 2 of an image ^N

When carrying out lifting wavelet transform, consider the situation that the border is expanded when actual.At this moment the data of selected image block will be up and down with about respectively expand two data.The coded sequence of cutting apart the back image block is for from left to right, from top to bottom.

So-called coefficient reorganization is exactly the sub-band coefficients each image block, according to the identical principle of location index, adds to one by one in the respective sub-bands of entire image.So that carry out the inverse transformation of Lifting Wavelet.

2. lifting wavelet transform

Native system adopts JPEG2000 to recommend the wavelet filter that uses, and promptly adopts Daubechies (9-7) filter when lossy compression method, and adopts Le Gall (5-3) wavelet filter during for lossless compress.And the simplification implementation algorithm of the lifting wavelet transform of employing this paper proposition.The number of transitions n of lifting wavelet transform generally is taken as 3, i.e. n=3.

3. quantification treatment

This algorithm adopts the lossy compression method pattern, and the floating data after the lifting wavelet transform is carried out quantification treatment.So that wavelet coefficient is mapped to integer field from the floating-point territory, simultaneously as much as possiblely eliminate human psycho-visual redundancy, obtain the compression ratio of maximum possible.

Human visual system (HVS) is revealed different visual sensitivities to the change list of different frequency bands.Be that HVS is responsive especially to the variation of low frequency signal, and insensitive for the variation of high-frequency signal; And HVS is relatively more responsive for the variation of the edge of horizontal direction and vertical direction and details, and insensitive for the variation to the details of angular direction.

Therefore, we can go out for different frequency bands according to these characteristics design of human visual system, the vector quantizer of different quantization steps.

If O (r) expression visual direction quantizing factor, expression formula is

$O\left(r\right)=2^{R_r},$ r＝0，1，2，3

R wherein _rThe index of representing the visual direction quantizing factor of different directions on the same level of resolution, and

R ₀+R ₁+R ₂+R ₃＝32

They take the space of 4 bytes altogether, when storage and transmission, flow to decoder prior to code stream.

S (k) expression vision band quantizing factor, expression formula is

S(k)＝2 ^k-1，k＝0，1，2，3

Wherein r represents the frequency band of different directions on the same level of resolution, and k represents level of resolution; We can set the quantization step Δ of the subband on the r direction on the k class resolution ratio grade thus _{K, r}For

${\mathrm{\Δ}}_{k,r}=\mathrm{\α}\×O\left(r\right)\×S\left(k\right)=\mathrm{\α}\×2^{R_r+k-1}$

Wherein α represents to quantize to adjust the factor; Quantize to adjust the factor and can carry out adaptive adjustment, for example can be set as follows according to the visual importance of current block b according to the characteristic of current coefficient block b:

$\mathrm{\α}=1+\underset{k=0}{\overset{4^{N-M}-1}{\mathrm{\Σ}}}{\mathrm{TB}}_k/(4^{N-M}\×{\mathrm{TB}}_b)$

In addition, for the sake of simplicity, can also directly to make α be a constant or directly get α=1.Then quantization operation is with the wavelet coefficient y of subband b _b(u v) is quantified as quantization parameter q _b(u, v):

The re-quantization process is,

y _b(u，v)＝q _b(u，v)×Δ _k，r。

4. adopt improved spiht algorithm:

Coefficient sets behind the entire image wavelet transformation just equals the lowest frequency subband and adds all details tree chunks.If the lowest frequency subband is also carried out piecemeal according to the size of details tree chunk, and the lowest frequency subband is regarded as top wavelet coefficient, then through behind 3 grades of wavelet transformations, whole wavelet coefficient just can be divided into 4 levels or 4 level of resolution.

(1) wavelet tree structure

If we change the tree structure of traditional spiht algorithm and use the tree structure that provides in the EZW algorithm instead.Our previously defined details tree group is with regard to many father's nodes like this.If still set the mode of the definition of chunk according to the front details, the definition tree chunk, then the coefficient sets behind the entire image wavelet transformation just equals wavelet coefficients set of all tree chunks.

Though, in our algorithm, still use the tree structure of EZW algorithm, difference is arranged on scan mode slightly.No longer use tree-like scan mode in EZW algorithm and the spiht algorithm, but all wavelet tree chunks are carried out level scanning.The former just likes the scanning of the preorder traversal pattern in the binary tree structure, and the latter similarly is the scanning of the level traversal formula in the binary tree.But, many wavelet tree chunks are arranged here, therefore, scan all tree root pieces earlier, and then scan the next level of all trees, promptly scan the coefficient block in the lowest frequency subband (or top) earlier.In same subband, then adopt the order of raster scan to scan with one deck.Scanning sequency as shown in Figure 3.

(2) piecemeal explanation

Because coefficient all has 4 son's coefficients in the adjacent high frequency band in each low-frequency band, therefore, in each low frequency sub-band,, be exactly according to 2M * 2M piecemeal in the then adjacent high-frequency sub-band according to M * M piecemeal, and except the lowest frequency subband at tree root place.Because the coefficient in the coefficient in the lowest frequency subband and next nearby frequency bands is one to one, both branch block sizes are duplicate.If carry out 3 grades of wavelet transformations, lowest frequency subband (promptly is the 0th class resolution ratio layer according to following definition) carries out piecemeal according to 2 * 2, and then the 1st class resolution ratio layer also just carries out piecemeal according to 2 * 2.And each later layer is just according to 2 ^kM * 2 ^kM carries out piecemeal.

(3) symbol description

If (i j) is a wavelet coefficient set of blocks to X, and (i j) is the piece index of this Wavelet Coefficient Blocks in the subband at its place.For positive integer n, note

$S_n\left(X(i,j)\right)=\left\{\begin{array}{cc}1,& \underset{c(u,v)\∈X(i,j)}{\max }\left\{\right|c(u,v)\left|\right\}\≥2^n\\ 0,& \mathrm{others}\end{array}\right.$

If S _n(X (i, j))=1, (i is j) about threshold value 2 then to claim X ⁿBe important; Otherwise claim that (i is unessential j) to X.

Use symbol

R=0,1,2,3; K=0,1,2 ..., m represents that level of resolution is k, the coordinate set of all the coefficient block left upper apex in the r direction subband,

Represent the coefficient coordinate set on the whole subband.Wherein m represents the number of times of wavelet decomposition; K=0 represents lowest band, other and the like; R=0 represents the LL subband, and r=1 represents the HL subband, and r=2 represents the LH subband, and r=3 represents the HH subband.When k=0, r can only equal 0, because have only a LL subband on the lowest band.When k ≠ 0, r ≠ 0, because except lowest band, other frequency bands all have the subband of 3 directions.

(4) ordered list

Because in this algorithm, employing be the level scan mode, and in the level each subband is adopted the blocked scan mode that is.Thereby in the ordered list that is provided with, the inessential subclass table of the LIS in traditional spiht algorithm just changes the inessential coefficient block table of LIB into, and inessential coefficient table of LIP and LSP significant coefficient table then still adopt.Each list item in three ordered lists all uses coordinate, and (u v) identifies.Coordinate (u, v) represent so that (u, v) coefficient is the coefficient block (this coordinate can transform mutually with the piece coordinate of this coefficient block) of left upper apex in LIB by expression coefficient coordinate in LIP and LSP.The size of coefficient block is by the decision of the resolution progression at this coefficient block place, and (u v) can determine the level of resolution and the subband direction at this coefficient place by the coefficient coordinate.

(5) algorithmic descriptions

Spiht algorithm key step after the improvement is as follows:

1) threshold value and ordered list initialization:

If threshold value T=2 ⁿ, wherein

Initialization k=0.Initialization LSP={} is an empty set, LIP={},

2) conspicuousness scanning is made up of following two big steps:

(1) checks that successively (u v), determines its whether significant coefficient to all wavelet coefficients among the LIP;

If important, then export 1 and sign bit, wherein the coding of sign is respectively 1 and 0, then will (i j) deletes from LIP, and adds the afterbody of LSP to.

If inessential, then export 0.

(2) each list item among the LIB is handled successively, is determined whether it is important:

If important, then export 1, enter then the piece index for (u, in the coefficient block v), according to from top to bottom, order from left to right successively each node in the processing block (k, l).The processing method of each node is as follows: if (k l) is significant coefficient, then exports 1 and sign bit, and adds it afterbody of LSP to; If inessential, then export 0, and add it afterbody of LIP to.

If inessential, then export 0.

After list item among the LIB before the ordering scanning beginning all handled, this minor sort scanning process finished.

3) fine scanning: to each list item among the LSP (u, v), if (u v) is not new interpolation the, then output in the scanning process of just carrying out | c _{I, j}| two n important position, wherein T=2 that advance in the expression ⁿIt is preset threshold in the scanning process.

4) make n=n-1, when n=0 jumps to 5), otherwise jump to step 2).

5) symbol of exporting in this conspicuousness scanning result is carried out arithmetic coding (can skip this step if do not carry out arithmetic coding).

6) result of the result of arithmetic coding and fine scanning is write code stream, and judge whether k equals 3, if then finish whole cataloged procedures; If not, then make k=k+1, and jump to 7).

7) make that LSP={} is an empty set, LIP={},

Wherein (u v) still carries out according to Fig. 4 (a) with (c) with putting in order of r among LIP and the LIB.And make T=2 ^m, n=m.Jump to 2 then).

5. arithmetic coding

The primitive rule of arithmetic coding is: the interval A=[0 of initialization codes at first, 1], encoded point C=0; Carry out interative computation according to following formula then:

$C=C+A\×P_{c_j};$ A＝A×p _j

P wherein _jWith

Be respectively the probability and the accumulated probability of current sign correspondence.

The arithmetic coding pattern has based on two kinds of the fixed mode of probability statistics and adaptive models.The former needs in advance the code word in the code book to be carried out prescan, and to add up the probability that each code word occurs, this process amount of calculation is bigger, and computational efficiency is not high; If the new not code word in code book occurs, then the image restoring effect will be undesirable.And the probability of each symbol is made as identical value when adopting the latter initial, changes its probable value accordingly according to the code word that occurs then, gets final product on only need adding in code book if new code word occurs.As long as encoder is used the method for identical initial value and identical change value, their probabilistic model will be consistent so.Encoder receives next code word it is encoded, and changes probabilistic model then, and decoder is according to current mode decoding, and then the probabilistic model of change oneself.

Binary arithmetic coding is a kind of arithmetic coding method the most commonly used in the reality, and the binary system of why saying so is because the character of its input has only two kinds.If include a plurality of characters in the information source character set, then, become string of binary characters earlier with a series of binary decision of these characters processes, carry out arithmetic coding again.

In 2 input characters of binary arithmetic coder, one that probability of occurrence is big is called MPS (More Probable Symbol), and another is called LPS (Less Probable Symbol).If the probability that LPS occurs is Q _eThen the probability of MPS appearance is 1-Q _eSymbol in symbols streams " 0 " and " 1 " corresponding respectively LPS or MPS.Along with the variation of the probability of symbol appearance in the symbols streams that is encoded, adaptive change also will take place in this corresponding relation.Its coding rule is as follows:

1) for MPS, encoded point: C=C; Between the code area: A=A (1-Q _e)=A-AQ _e

2) for LPS, encoded point: C=C+A (1-Q _e)=C+A-AQ _eBetween the code area: A=AQ _e

In Huffman coding, the shortest code word is 1bit, so even to the symbolic coding of probability of occurrence maximum the time, increase 1bit again on the code stream that also needs to have finished in front.And in arithmetic coding, the MPS coding not being increased the code stream length of having finished, this is that arithmetic coding is than the superior place of Huffman coding.

When the specific implementation binary arithmetic coding, resolve following problem:

1) in the process of constantly dividing between the code area, interval width A is more and more littler, is used for then representing that the digit of A is more and more;

Need use the higher multiplying of cost when 2) finishing encoded point calculating and interval cutting apart;

3) a plurality of 1 the time when occurring continuously in the code stream of having finished, if in the end add 1 on one in the next code process, will continuously change the code word that has finished the front, produce continuously a plurality of 0, up to 1 occur till.

In order effectively to realize the arithmetic coding algorithm, people have proposed many solutions, construct dissimilar arithmetic encoders.The present invention will adopt the solution that provides in the JPEG2000 standard.

Come the interval A of calculation code with the arithmetical operation of limited precision, be unlikely to increase with the increase of code word string with the figure place of the significant digits that guarantee A.For example, A=0.001 can be with 1.0 * 2 ^-3Represent, when A is split into littler value, then increase exponential quantity, and the figure place of significant digits keeps within the limits prescribed.This scope is chosen between 0.75～1.5 among the JPEG2000, when A＜0.75, A be multiply by 2, and significant digits are remained between 0.75～1.5, so still can represent with original figure place, and these codings that all help software and hardware are realized.This process is called as normalization again.Take advantage of 2 computings to realize with shift left operation to register.Behind interval A normalization again, also normalization thereupon of code word string C.When 0.75≤A≤1.5, AQ _e≈ Q _eUtilize this approximate expression, encoded point calculating and interval can be cut apart simplification, and need not to do multiplication, about the problem of carry, can adopt " bit is filled (Bit Stuffing) " technology to avoid.The bit filling technique that adopts among this technology and the JPEG is similar.

If above-mentioned adaptive model is used in the binary arithmetic coding, just constituted adaptive binary arithmetic coding.

With above-mentioned foundation desirable embodiment of the present invention is enlightenment, and by above-mentioned description, the related work personnel can carry out various change and modification fully in the scope that does not depart from this invention technological thought.The technical scope of this invention is not limited to the content on the specification, must determine its technical scope according to the claim scope.

Claims (4)

1. one kind based on improving the method that wavelet-transform image compression method realizes mobile phone mobile portal, it is characterized in that comprise coded portion and decoded portion, concrete steps are as follows:

Coded portion is made up of wavelet transformation, quantification and three basic modules of entropy coding;

Described wavelet transformation refers to carry out earlier image segmentation, and original image is carried out piecemeal, carries out the coefficient reorganization again, the sub-band coefficients of each image block, according to the identical principle of location index, add to one by one in the respective sub-bands of entire image, then to the independent lifting wavelet transform of each image block;

Described quantification refers to the floating data after the lifting wavelet transform is carried out quantification treatment, so that wavelet coefficient is mapped to integer field from the floating-point territory;

Described entropy coding refers to that by improved spiht algorithm and arithmetic coding quantized transform coefficients being changed into one is used for transmitting or the compressed bit stream of storing, and writes bit stream to the encoding stream of all images piece one by one according to orderly scanning sequency then;

Described improved spiht algorithm adopts the tree structure of EZW algorithm, and all wavelet tree chunks are carried out level scanning, has scanned all tree root pieces earlier, scans the next level of all trees again; The perhaps first coefficient block in the scanning lowest frequency subband, the order that adopts raster scan again is to scanning in the same subband of one deck; Perhaps scan the coefficient block in the high-frequency sub-band earlier, the order that adopts raster scan again is to scanning in the same subband of one deck;

In decoding end, after the data of each image block in the bit stream are carried out entropy decoding, concern, write the relevant position of entire image one by one according to the wavelet coefficient and the position between the entire image wavelet coefficient of image block, carry out re-quantization and contrary lifting wavelet transform then, just can recover image.

2. according to claim 1 based on improving the method that wavelet-transform image compression method realizes mobile phone mobile portal, it is characterized in that, each index block in the described original image is through behind wavelet transformation, and its each subband index piece lays respectively on the identical index position of respective sub-bands behind the whole original image wavelet transformation.

3. according to claim 1 based on improving the method that wavelet-transform image compression method realizes mobile phone mobile portal, it is characterized in that the number of transitions n of described lifting wavelet transform 〉=3.

4. according to claim 1 based on improving the method that wavelet-transform image compression method realizes mobile phone mobile portal, it is characterized in that, described quantification is that the floating data after the lifting wavelet transform is carried out quantification treatment, so that wavelet coefficient is mapped to integer field from the floating-point territory;

If O (r) expression visual direction quantizing factor, expression formula is

$O\left(r\right)=2^{R_r},$ r＝0，1，2，3

R wherein _rThe index of representing the visual direction quantizing factor of different directions on the same level of resolution, and

R ₀+R ₁+R ₂+R ₃＝32

They take the space of 4 bytes altogether, when storage and transmission, flow to decoder prior to code stream;

S (k) expression vision band quantizing factor, expression formula is

S(k)＝2 ^k-1，k＝0，1，2，3

Wherein r represents the frequency band of different directions on the same level of resolution, and k represents level of resolution; The quantization step Δ of the subband on the r direction on the k class resolution ratio grade _{K, r}For

${\mathrm{\Δ}}_{k,r}=\mathrm{\α}\×O\left(r\right)\×S\left(k\right)=\mathrm{\α}\×2^{R_r+k-1}$

Wherein α represents to quantize to adjust the factor; Quantize to adjust the factor and can carry out adaptive adjustment according to the characteristic of current coefficient block b.

Images