CN102547287A - Raptor code-based image coding method - Google Patents

Abstract

The invention provides a Raptor code-based image coding method, mainly overcoming the defect that in the traditional encoding of the combined signal source and channel, the code rate is fixed so as to enable the channel utilization rate to be low. The realization process comprises the following steps of: 1, carrying out five-stage 9/7 wavelet transformation for an original image to produce sub-bands of different frequencies, and interweaving coefficients in the sub-bands; 2, carrying out plane encoding for the sub-bands one by one according to the sequence from low frequency to high frequency, and combining encoded code streams into a plurality of packets corresponding to the sub-bands; 3, carrying out Raptor encoding for the code stream in each packet to obtain the code stream to be transmitted and with corrected code information; 4, transmitting the code streams to be transmitted through a binary deletion channel at the transmitter end, and receiving the transmitted code stream information at the receiver end; and 5, decoding the received code stream information according to the sequence opposite to the sequence at the transmitter end, and finally obtaining a reconstructed image. The invention has the advantage of low complexity, so that the progressive image compression and transmission with strict real-time requirements can be realized.

Description

Method for encoding images based on the Raptor sign indicating number

Technical field

The invention belongs to technical field of image processing, particularly relate to method for compressing image, can be used for complexity and the strict gradual image compression transmission of real-time.

Background technology

Image is one of topmost instantly information recording method formula, and especially along with broad application such as video telephone, HDTV, video conference, remotely sensed image and remote monitoring, image has become the main carrier of information interchange in people's life.Yet,, image receives interference of noise easily when transmitting through wireless channel, and the communication quality of image can descend greatly, so the reliable transmission of image in wireless channel becomes the focus of research instantly.

In recent years; Many scholars have proposed the joint source-channel coding scheme; Optimized design is carried out in the source encoding and the channel combined consideration of wireless communication system, come the reliability of abundant guarantee information transmission, under the situation of identical complexity, can obtain good systematic function.In the chnnel coding process,, come raw information is protected through on the raw information basis, adding redundant information.The existing channel Methods for Coding has its intrinsic defective: the code check of coding is fixed; Thereby the joint source-channel coding based on them also has deficiency: the code check of whole system is fixed; The structure of sign indicating number was confirmed before message transmission, can not the abundant coded identification of online generation be deciphered raw information.When receiving terminal can not correct decoding raw information, coded message can only resend, and the occupancy that this can increase channel lowers efficiency.

Summary of the invention

The objective of the invention is to overcome the shortcoming and defect of above-mentioned prior art, propose a kind of method for encoding images, to improve the utilization ratio of channel based on the Raptor sign indicating number.

For realizing above-mentioned purpose, performing step of the present invention comprises as follows:

(1) original image is carried out Pyatyi 9/7 wavelet transformation, produce the subband of different frequency, and the coefficient in the subband is interweaved, obtain having the fault-tolerant changing image coefficient of ability;

(2) method that the changing image coefficients by using is rounded up quantizes;

(3), each subband is carried out Bit-Plane Encoding one by one, and the code stream after will encoding is formed a packet, the corresponding subband of each packet according to the order from the low frequency to the high frequency;

(4) code stream in each packet is carried out Raptor coding, the code stream to be sent that obtains having error correction code information;

(5), code stream to be sent is transmitted the code stream information after receiving terminal receives transmission through binary erasure channel at transmitting terminal;

(6) code stream information that receives the termination receipts is carried out Raptor decoding, the described packet information of recovering step (3);

(7) the data package informatin is carried out the bit plane decoding, recover the image transform coefficient, and this coefficient is carried out deinterleaving;

(8) the image transform coefficient after the deinterleaving is carried out 9/7 wavelet inverse transformation, obtain final decoded picture.

The present invention compared with prior art has following advantage:

The present invention is owing to adopt the Raptor sign indicating number with no code check characteristic; Can not only remedy the defective of the code check fixed form of traditional joint source-channel coding; And can online generation abundant coded identification has improved the utilization ratio of channel so that raw information is deciphered; Simultaneously, owing to having adopted the Bit-Plane Encoding mode and the coefficient in the subband behind the wavelet transformation being interweaved, so can under the situation of road inclement condition, still can obtain high-quality recovery image; The present invention has the low advantage of complexity in addition, can carry out the strict gradual image compression transmission of real-time.

Description of drawings

Fig. 1 is a general flow chart of the present invention;

Fig. 2 is a sub-band coefficients intersection chart of the present invention;

Fig. 3 is packet structure figure before the Raptor coding;

Fig. 4 is Raptor coding back packet structure figure;

Fig. 5 is a binary erasure channel illustraton of model of the present invention;

Fig. 6 is the generator matrix sketch map of the present invention when encoding;

Fig. 7 is the image that emulation of the present invention is adopted.

Embodiment

With reference to Fig. 1, concrete steps of the present invention are following:

Step 1 is carried out Pyatyi 9/7 wavelet transformation to original image, produces the subband of different frequency, and the coefficient in the subband is interweaved, and obtains having the fault-tolerant changing image coefficient of ability.

With reference to Fig. 2, the step that interweaves of the coefficient in the subband is following:

1a) subband was carried out four fens, produce four equal piece M ₁, M ₂, M ₃, M ₄, shown in Fig. 2 (a);

1b) with M ₁Piece is divided into four equal pieces, and according to M ₁The position at place intersects and places these four equal pieces, shown in Fig. 2 (b);

1c) according to 1b) described process, with M ₂, M ₃, M ₄Each piece is divided into four equal pieces respectively, and according to their position, intersects and place the position of four equal pieces separately.

Step 2; The method that the changing image coefficients by using is rounded up quantizes; The sign bit that is the changing image coefficient is constant; Add 1 as the value after quantizing if the value of the fractional part of this coefficient, is then got the value of the integer part of this coefficient more than or equal to 0.5, otherwise the integer part of directly getting this coefficient is as the value after quantizing.

Step 3 according to the order from the low frequency to the high frequency, is carried out Bit-Plane Encoding to each subband one by one, and the code stream after will encoding is formed a packet, the corresponding subband of each packet.

The step of said Bit-Plane Encoding is following:

3a) according to the maximum M of the coefficient in the subband; Confirm the highest order Plane n, and

3b) coding highest order Plane n, and the threshold value of calculating this highest order plane is 2 ⁿIf the coefficient in the subband is then exported 1, otherwise is exported 0 greater than this threshold value;

3c) coding time high bit plane, the n value subtracts 1, and upgrading threshold value is 2 ^N-1If the coefficient in the subband is then exported 1, otherwise exported 0 greater than this threshold value, repeating this process is 0 o'clock up to the n value, encoded bitplane 0.

Code stream behind each subband (Wavelet Band) Bit-Plane Encoding can be formed a packet, and its structure is as shown in Figure 3, is made up of start mark byte (BeginByte), data byte sum (DataNumber) and information source information (SourceData).

Step 4 is carried out Raptor coding, the code stream to be sent that obtains having error correction code information to code stream in each packet.

The Raptor coding is the key technology that realizes the object of the invention, and it is made up of two links: precoding and LT coding, and the precoding here is made up of low-density check code LDPC coding and Half coding; In the Raptor cataloged procedure, need to calculate earlier intermediate symbols, this intermediate symbols is the prerequisite of carrying out the LT coding, the LT coding carries out the process of XOR just to middle symbol; Code stream in each packet carries out after the Raptor coding; Had error correction code information; Its structure is as shown in Figure 4, is made up of start mark byte (BeginByte), data byte sum (DataNumber), information source information (SourceData) and Raptor coded message (RaptorData).

The concrete steps of Raptor coding are following:

4a) according to the source symbol length K, calculate LDPC coded identification number S, Half coded identification number H and intermediate symbols length L respectively:

For S, find the solution the smallest positive integral X that satisfies X * (X-1) >=2 * K earlier, then S be satisfy S >=ceil (prime number of the minimum of 0.01 * K)+X, wherein ceil (x) expression is more than or equal to the smallest positive integral of x;

For H, it is the smallest positive integral that satisfies choose (H, ceil (H/2)) >=K+S, wherein, choose (i, j) number of combinations of j element is selected in expression from i element;

For L, it is K, S, H sum, i.e. L=K+S+H.

4b) according to the precoding relation, structure generator matrix A:

4b1) setup parameter and clear and definite precoding relation

If C [0] ..., C [L-1] representes L intermediate symbols, C [0] then ..., K intermediate symbols before C [K-1] expression, C [K] ... C [K+S-1] representes S LDPC coded identification, C [K+S] ... C [L-1] representes H Half coded identification.Precoding relation is the relation between L-K the symbol of preceding K symbol and back of intermediate symbols, and it comprises two layers of meaning, promptly preceding K intermediate symbols and the relation of S LDPC coded identification and the relation of a preceding K+S intermediate symbols and H Half coded identification;

4b2) calculate LDPC encoder matrix G_LDPC

G_LDPC is a S * kth moment battle array, and it is the ground floor implication of precoding, expression LDPC cataloged procedure, and G_LDPC can calculate through the relation of a preceding K intermediate symbols and S LDPC coded identification.

In document RFC5053 " Raptor Forward Error Correction Scheme for Object Delivery ", K intermediate symbols and S LDPC coded identification satisfy following relation before providing:

I does following the processing respectively since 0 to K-1:

a＝1+floor(i/S)％(S-1)

b＝i％S

C[K+b]＝C[K+b]^C[i]

b＝(b+a)％S

C[K+b]＝C[K+b]^C[i]

b＝(b+a)％S

C[K+b]＝C[K+b]^C[i]

Wherein, a, b, i representes integer, and " ^ " representes XOR, and floor (x) expression is smaller or equal to the smallest positive integral of x;

4b3) calculate Half encoder matrix G_Half

G_Half is the rank matrix of a H * (K+S); It is the second layer implication of precoding; Expression Half cataloged procedure; G_Half can calculate through the relation of a preceding K+S intermediate symbols and H Half coded identification, is the relation of K+S intermediate symbols before representing and H Half coded identification, at first does a definition:

g[i]＝i^floor(i/2)，

The g here [i] is a gray code sequence; Be to have only 1 bit difference in the sequence between former and later two code words, the subsequence with m [H '] expression g [i] has 1 of the individual bit of H ' in the binary representation of each code word of this subsequence; And H '=ceil (H/2); And with j element in m [j, H '] expression m [the H '] sequence, then before K+S intermediate symbols concern as follows with H Half coded identification is satisfied:

If h is since 0 to H-1, j is since 0 to K+S-1, if the h bit of m [j, H '] binary representation equals 1, and C [h+K+S]=C [h+K+S] ^C [j] then, on the contrary constant, wherein, and h, j representes integer, " ^ " representes XOR;

4b4) calculate LT encoder matrix G_LT, and structure generator matrix A

G_LT is a K * L rank matrix, and it calculates according to the LT cataloged procedure that provides in the coding of the LT in 3GPP TS 26.346 standards.As shown in Figure 6; Generator matrix A is a L * L rank square formation; It is made up of LDPC encoder matrix G_LDPC, Half encoder matrix G_Half, LT encoder matrix G_LT, matrix I_S, matrix I_H and matrix 0_S * H, and wherein, matrix I_S is a S * S rank unit matrix; Matrix I_H is a H * H rank unit matrix, and matrix 0_S * H is a S * H rank null matrix;

4c) calculate the inverse matrix of generator matrix A, and calculate intermediate symbols C [0]～C [L-1]

Because generator matrix A is a square formation, therefore adopt full pivoting Gaussian reduction to calculate the inverse matrix A of generator matrix A ^-1, obtain inverse matrix A ^-1After, calculate intermediate symbols C [0]～C [L-1] according to following relation:

C＝A ^-1×D 1)

Wherein, C representes by L intermediate symbols C [0], C [1] ..., the column vector that C [L-1] forms, D representes the column vector be made up of L symbol, and its preceding S+H symbol is 0, back K symbology source symbol.According to relational expression 1), with the inverse matrix A of generator matrix A ^-1Carry out matrix multiple with column vector D, C calculates with column vector, can obtain intermediate symbols C [0], C [1] ..., C [L-1];

4d) utilize intermediate symbols C [the 0]～C [L-1] that calculates, it is carried out the LT coding, obtain Raptor coding back symbol.

Step 5 at transmitting terminal, is transmitted code stream to be sent through binary erasure channel, the code stream information after receiving terminal receives transmission.

The model of binary erasure channel is as shown in Figure 5, and binary message 0/1 arrives receiving terminal with the probability correct transmission of 1-p, and loses in channel with the probability of p, and wherein p representes the information dropout rate, and the E representative information is lost.

Step 6 is carried out Raptor decoding, recovering step 3 described packet informations to the code stream information that receives the termination receipts.

6a) according to the code stream information that receives, calculate the symbol lengths N and decoding length M that receive respectively, M=S+H+N, wherein S is a LDPC coded identification number, H is a Half coded identification number;

6b), construct decoding matrix B according to the building process of coding generator matrix A

Decoding matrix B is a M * L rank matrix, can construct according to the building process of coding generator matrix A, and promptly the preceding S+H of decoding matrix B is capable identical with the A matrix, and back N is capable to be calculated according to the LT cataloged procedure in 3GPP TS 26.346 standards;

6c) structure decoding equation

If C representes by L intermediate symbols C [0], C [1] ...; The column vector that C [L-1] forms; G representes the column vector be made up of M symbol, and its preceding S+H symbol is 0, and a back N symbol is the coded identification that receives; C regards the unknown number column vector as with the intermediate symbols column vector, and then decoding matrix B, intermediate symbols column vector C and column vector G formation decoding equation are:

B×C＝G

2)

6d) calculate intermediate symbols C [0]～C [L-1], obtain information source information through decoding again

Utilize the full pivoting Gaussian reduction of galois field solving equation 2); Calculate intermediate symbols vector C; After obtaining intermediate symbols column vector C, the LT encoder matrix G_LT and the intermediate symbols column vector C that will when coding, produce carry out matrix multiple, can decipher obtaining information source information.

Step 7 is carried out the bit plane decoding to the data package informatin, recovers the image transform coefficient, and this coefficient is carried out deinterleaving.

Said bit plane decoding concrete steps are following:

7a) corresponding subband of each packet, the coefficient initialization in the subband is 0, begins decoding from the highest order Plane n, calculates the threshold value 2 on this highest order plane ⁿIf, be input as 1, then coefficient and this threshold value addition in the subband, otherwise the coefficient in the subband is constant;

7b) decoding time high bit plane, the n value subtracts 1, and upgrading threshold value is 2 ^N-1If, be input as 1, then coefficient and this threshold value addition in the subband, otherwise the coefficient in the subband is constant;

7c) repetition 7b) be 0 o'clock up to the n value, decoded bit plane 0.

Step 8 is carried out 9/7 wavelet inverse transformation to the image transform coefficient after the deinterleaving, obtains final decoded picture.

Effect of the present invention further specifies through following concrete experimental data.

1. experiment condition

Experiment of the present invention is that to get size be 512 * 512, and gray scale is four width of cloth natural images such as Lena, Barbara, Goldhill and Munar of 8 bits, and is as shown in Figure 7; Wherein Fig. 7 (a) is Lena figure; Fig. 7 (b) is Barbara figure, and Fig. 7 (c) is Goldhill figure, and Fig. 7 (d) is Munar figure.

2. experiment content

Step according to Code And Decode of the present invention is carried out Code And Decode respectively to four width of cloth natural images shown in Figure 7.If source symbol length is 40; Coded identification length is 60, and the length of each symbol is 3 bytes, and the Loss Rate of binary erasure channel is 0.001; Be to calculate Y-PSNR PSNR under 0.125,0.25,0.5 and 1 the condition at code check bpp; And under square one, contrast the traditional chnnel coding RS coding and the PSNR value of method of the present invention, see table 1.

2. experiment comparative result

The present invention adopts the index of Y-PSNR PSNR as the numeral contrast, and it defines as follows:

$\mathrm{PSNR}=10\lg ({255}^2/[\underset{i=1}{\overset{I}{\mathrm{\Σ}}}\underset{j=1}{\overset{J}{\mathrm{\Σ}}}{(I_{\mathrm{ij}}-I_{\mathrm{ij}}^e)}^2/\mathrm{IJ}\left]\right)---3)$

Wherein, I, the size of J presentation video is 512 here, I _Ij ^eExpression original image pixels value, I _IjImage pixel value after expression decoding recovers.

The Y-PSNR PSNR performance of table 1 distinct methods is (unit: dB) relatively

Visible from table 1, the effect that the present invention is based on the Raptor coding is better than the effect of traditional RS coding.

Claims (6)

1. the method for encoding images based on the Raptor sign indicating number comprises the steps:

(1) original image is carried out Pyatyi 9/7 wavelet transformation, produce the subband of different frequency, and the coefficient in the subband is interweaved, obtain having the fault-tolerant changing image coefficient of ability;

(2) method that the changing image coefficients by using is rounded up quantizes;

(3), each subband is carried out Bit-Plane Encoding one by one, and the code stream after will encoding is formed a packet, the corresponding subband of each packet according to the order from the low frequency to the high frequency;

(4) code stream in each packet is carried out Raptor coding, the code stream to be sent that obtains having error correction code information;

(5), code stream to be sent is transmitted the code stream information after receiving terminal receives transmission through binary erasure channel at transmitting terminal;

(6) code stream information that receives the termination receipts is carried out Raptor decoding, the described packet information of recovering step (3);

(7) the data package informatin is carried out the bit plane decoding, recover the image transform coefficient, and this coefficient is carried out deinterleaving;

(8) the image transform coefficient after the deinterleaving is carried out 9/7 wavelet inverse transformation, obtain final decoded picture.

2. the method for encoding images based on the Raptor sign indicating number according to claim 1, wherein step (1) is described interweaves to the coefficient in the subband, carries out as follows:

2a) subband was carried out four fens, produce four equal piece M ₁, M ₂, M ₃, M ₄

2b) with M ₁Piece is divided into four equal pieces, and according to M ₁The position at place intersects and places these four equal pieces;

2c) according to 2b) described process, with M ₂, M ₃, M ₄Each piece is divided into four equal pieces respectively, and according to their position, intersects and place the position of four equal pieces separately.

3. the method for encoding images based on the Raptor sign indicating number according to claim 1, wherein step (3) is described carries out Bit-Plane Encoding one by one to each subband, carries out according to following steps:

3a) according to the maximum M of the coefficient in the subband; Confirm the highest order Plane n, and

3b) coding highest order Plane n, and the threshold value of calculating this highest order plane is 2 ⁿIf the coefficient in the subband is then exported 1, otherwise is exported 0 greater than this threshold value;

3c) coding time high bit plane, the n value subtracts 1, and upgrading threshold value is 2 ^N-1If the coefficient in the subband is then exported 1, otherwise exported 0 greater than this threshold value, repeating this process is 0 o'clock up to the n value, encoded bitplane 0.

4. the method for encoding images based on the Raptor sign indicating number according to claim 1, wherein step (4) is described carries out the Raptor coding to the code stream in each packet, carries out as follows:

4a) according to the source symbol length K, calculate LDPC coded identification number S, Half coded identification number H and intermediate symbols length L respectively;

4b) according to the precoding relation, structure generator matrix A;

4c) calculate the inverse matrix of generator matrix A, and calculate intermediate symbols C [0]～C [L-1];

4d) utilize intermediate symbols C [the 0]～C [L-1] that calculates, it is carried out the LT coding, obtain Raptor coding back symbol.

5. the method for encoding images based on the Raptor sign indicating number according to claim 1, wherein step (6) is described carries out Raptor decoding to the code stream information that receives the termination receipts, carries out as follows:

5a) according to the code stream information that receives, calculate the symbol lengths N and decoding length M that receive respectively, M=S+H+N, wherein S is a LDPC coded identification number, H is a Half coded identification number;

5b), make up decoding matrix B according to the building process of coding generator matrix A;

5c) structure decoding equation;

5d) calculate intermediate symbols C [0]～C [L-1], obtain information source information through decoding again.

6. the method for encoding images based on the Raptor sign indicating number according to claim 1, wherein step (7) is described carries out the bit plane decoding to the data package informatin, carries out according to following steps:

6a) corresponding subband of each packet, the coefficient initialization in the subband is 0, begins decoding from the highest order Plane n, calculates the threshold value 2 on this highest order plane ⁿIf, be input as 1, then coefficient and this threshold value addition in the subband, otherwise constant;

6b) decoding time high bit plane, the n value subtracts 1, and upgrading threshold value is 2 ^N-1If, be input as 1, then coefficient and this threshold value addition in the subband, otherwise the coefficient in the subband is constant;

6c) repetition 6b) be 0 o'clock up to the n value, decoded bit plane 0.

Images