CN103002280A - Distributed encoding/decoding method and system based on HVS/ROI (human vision system and region of interest) - Google Patents

Abstract

The invention relates to distributed encoding method and system based on HVS/ROI (human vision system and region of interest) and belongs to distributed encoding methods and systems. The system comprises an encoder and a decoder. The encoder divides an input video Wyner-Ziv frame into macroblocks C and macroblocks AC by an HVS JND (just noticeable distortion) model, and macroblocks of a K frame and macroblocks of the WZ frame are encoded independently. The decoder is used for joint decoding of received code streams. The HVS JND model is introduced, so that data processing and transmission of JND thresholds in next region are reduced, transmission rate is reduced, and energy consumption and bandwidth are economized. A glare region with high movement violence and brightness contrast is subjected to ROI extraction and is subjected to entropy encoding, and subjective quality is improved while encoding complexity is not increased as far as possible.

Description

Distributed decoding method and system based on HVS ROI

Technical field

The present invention relates to a kind of distributed coding method and system, particularly a kind of based on HVS﹠amp; Distributed decoding method and the system of ROI.

Background technology

Along with the development of sensor technology, underground communica tion technology and technology of Internet of things, for performance and the security reliability thereof that improves the coal mine underground monitoring system provides technical guarantee.Rely on project of national nature science fund project " the mine multimedia disaster communication system-based theoretical research of long distance W MSN " to propose, intend to adopt have self-organization, place flexibly, collection transmission work that wireless multimedia sensor network (Wireless Multimedia Sensor Network, WMSN) that mobility is strong is realized monitoring image after the mine disaster.

In WMSN, the video sensing node of computing capability and energy constraint has two basic demands to encoder: (1) is subjected to joint behavior, energy limited, and encoder should have the characteristics of low complex degree, low-power consumption; (2) be subjected to speed, limit bandwidth, encoder should have high compression rate.Traditional multimedia video compression and coding standard, the 5-10 that the common Video coding end of its complexity is decoding end doubly is not suitable for WMSN.And distributed video coding (Distributed Video Coding, DVC) has following characteristics: (1) encoder complexity is relatively low, and decoding complex degree is relatively high; (2) wireless channel that is prone to error code, packet loss had good robustness; (3) compression ratio is higher, and easily realizes multilevel coding.DVC makes the complexity of coding side transfer to decoding end by intraframe coding and interframe decoding.Very suitable being applied to requires the wireless video occasions such as harsh mobile video telephone, wireless camera machine monitoring to encoder complexity and energy consumption.

Do not have natural light irradiation under the coal mine, can only lean on artificial light, and subsurface environment is complicated, can't guarantee that All Ranges can both lay the installation lighting apparatus, most of zone, down-hole belongs to the half-light environment, in some tunnel even do not have light.But, by the mine lamp direct irradiation or shine reflective object, will cause visual field internal space or temporal extreme brightness contrast, and cause the glare vision that visual discomfort and object visibility reduce.Mostly existing DVC system model is that coding side independently carries out Wyner-Ziv and intraframe coding, utilize side information and virtual correlated channels combined decoding in decoding end, and the compression of only characteristic on the space of video own and temporal redundancy and the transform domain being carried out is processed, and does not consider the specific photographed scene of video and visual effect.

Summary of the invention

The object of the present invention is to provide a kind of subjective quality that improves, reduce transmission rate, save a kind of based on HVS﹠amp of energy consumption and bandwidth; The distributed coding system of ROI.

For achieving the above object, provided by the invention based on HVS﹠amp; The distributed coding method of ROI comprises Code And Decode, and wherein, the coding method step is as follows:

A. input video frame is divided into key frame and Wyner-Ziv frame, key frame is carried out intraframe coding, described key frame is the K frame, and H.264 intra encoder is carried out intraframe coding to key frame; The Wyner-Ziv frame is carried out HVS﹠amp; ROI Wyner-Ziv coding;

B. described HVS﹠amp; The method of ROI Wyner-Ziv coding comprises: the JND model by HVS is divided into C macro block and AC macro block with the Wyner-Ziv frame;

Described JND model method comprises:

B1. the Wyner-Ziv frame is divided into 8 * 8 piecemeals, each piecemeal is the B piece;

B2. calculate the JND threshold value of each pixel in the B piece according to following formula:

JND(x，y，t)＝f(idl(x，y，t))JND ^s(x，y)

Wherein, luminance errors value between the adjacent t moment of idl (x, y, t) expression and t-1 average frame constantly, JND ^sJND is in the spatial domain threshold value in (x, y) expression;

$\mathrm{idl}(x,y,t)=\frac{1}{2}(I(x,y,t)-I(x,y,t-1)+\stackrel{\‾}{I}(x,y,t)-\stackrel{\‾}{I}(x,y,t-1))$

JND ^s(x，y)＝T ^l(x，y)+T _Y ^t(x，y)-C _Y×Min{T ^l(x，y)，T _Y ^t(x，y)}

Wherein, T ^l(x, y) expression background luminance adapts to influence function, T _Y ^tHVS is stronger than the sensitiveness of texture compact district for level and smooth district for the reaction of (x, y) expression texture influence function, pixel corresponding coordinate in (x, y) presentation video, C _YCoefficient correlation between expression background luminance and two kinds of influencing factors of texture masking;

B3. calculate forward, backward and the mean pixel SAD of current block B piece according to following formula:

$\mathrm{SAD}=\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}|w(x,y)-r(x,y)|$

Wherein, w (x, y) and r (x, y) represent respectively the pixel value that encoding block and reference block are located at coordinate (x, y) in the WZ frame;

B4. obtain coding side side information ESI by following formula:

ESI＝Min _SAD{FB，BB，AB}

Wherein, FB is forward direction consecutive frame corresponding blocks, and FB is the abbreviation of Forward Block; BB is backward consecutive frame corresponding blocks, and BB is the abbreviation of Backward Block; AB is the corresponding average block of front and back consecutive frame, and AB is the abbreviation of Average Block;

B5. calculate respectively consecutive frame SAD _BIn maximum

And minimum value

Note

Obtain respectively t＜SAD _BWith t＞SAD _BThe time pixel ratio ρ ₀, ρ ₁With average gradient value m ₀, m ₁, then the overall average Grad is: m=ρ ₀m ₀+ ρ ₁m ₁

B6. the variance of calculating prospect and background is:

v＝ρ ₀ρ ₁(m ₀-m ₁) ²；

B7. exist Traversal v finds out maximum v on the interval _Max, v _MaxCorresponding t is in prospect and the background variance maximum, and this moment, the t value was threshold value T _ROI

B8. according to the T that obtains _ROIValue tentatively extracts foreground image, by comparing SAD _BWith T _ROIValue extract foreground image, for greater than T _ROIThen the extracting section of value is carried out foreground image one time to b8 by step b5 more out as foreground image, obtains more accurate secondary T _ROI

B9. calculate the predicted distortion of each point according to the distortion function D (x, y) of ESI, obtained the predicted distortion value of each point by following formula:

D (x, y)=| ESI (x, y)-I (x, y) | wherein, ESI (x, y) and I (x, y) represent respectively ESI value and the pixel value that coordinate (x, y) is located in the WZ frame.

B10. according to following formula piece B being carried out JND differentiates, divide AC piece and C piece, become the C piece for distortion value less than or equal to the macroblock partitions of JND threshold value, and be the AC piece for distortion value greater than the macroblock partitions of JND threshold value, and obtain the two-value mask Block that this WZ two field picture piece is divided _Mask:

${\mathrm{Block}}_{\mathrm{mask}}=\left\{\begin{array}{cc}1& \left(\mathrm{\&Sigma;}{\mathrm{JND}}_{\mathrm{mask}}\right)\&GreaterEqual;\mathrm{\&epsiv;}\&CenterDot;\mathrm{Num}\\ 0& \left(\mathrm{\&Sigma;}{\mathrm{JND}}_{\mathrm{mask}}\right)<\mathrm{\&epsiv;}\&CenterDot;\mathrm{Num}\end{array}\right.$ ${\mathrm{JND}}_{\mathrm{mask}}=\left\{\begin{array}{cc}1& (\mathrm{JND}(x,y)\&GreaterEqual;D(x,y))\\ 0& (\mathrm{JND}(x,y)<D(x,y))\end{array}\right.$

Wherein, Num is the total quantity of pixel in each macro block, and ε gets 0.1.

B11. the AC piece is done DCT and 2 ^MThe level unified quantization, the extraction coefficient band also carries out Zigzag scanning, according to selected quantization matrix conversion coefficient is carried out uniform quantization; And according to the SAD of following formula calculating K frame and WZ frame same position AC piece gradient _B

${\mathrm{SAD}}_B=\underset{(j,k)\&Element;B}{\mathrm{\&Sigma;}}\left|G_K\right[F(j,k)]-G_W[F(j,k)\left]\right|$

Wherein K represents that key frame, W are that Wyner-Ziv frame, B are single macro block, the pixel value that F (j, k) locates for z coordinate (j, k).

B12. adopt following formula ROI macro block decision criteria, obtain region of interest two-value mask ROI _Mask:

${\mathrm{ROI}}_{\mathrm{mask}}=\left\{\begin{array}{cc}1& ({\mathrm{SAD}}_B\&GreaterEqual;T_{\mathrm{ROI}})\\ 0& ({\mathrm{SAD}}_B<T_{\mathrm{ROI}})\end{array}\right.$

Wherein, ROI _MaskBe the two-value mask of region of interest, as macro block gradient SAD _BThen be set to 0 less than gradient T, otherwise be 1; To two-value mask ROI _MaskCarry out simple morphology and process, first corrosion and then expansion make contour of object level and smooth, extract ROI by following formula again:

$\mathrm{ROI}=\left\{\begin{array}{cc}{W}_B& (W_B\&SubsetEqual;{\mathrm{ROI}}_{\mathrm{mask}})\\ 0& (W_B\&NotSubset;{\mathrm{ROI}}_{\mathrm{mask}})\end{array}\right.$

Wherein, W _BRepresent corresponding AC macro block in the current WZ frame, if W _BCorresponding ROI _MaskBe 1 and be the ROI piece, be non-ROI piece otherwise set to 0, finally obtain the ROI region of interest of present frame;

B13. to Block _Mask, ROI _MaskAnd the ROI piece after quantizing carries out the Ha Fuman compression coding; Non-ROI piece extraction bit plane to the AC macro block carries out the LDPC coding, if the image sets of present encoding is last image sets, then withdraws from HVS﹠amp; The ROI distributed video coding;

C. the AC macro block is extracted ROI, the label information of ROI macro block and each macro block is carried out the entropy coding, non-ROI macro block is carried out encoding based on the Wyner-Ziv of LDPC (the LDPC here can quote top definition);

D. the concrete steps of described coding/decoding method are:

D1. H.264 intraframe decoder is obtained key frame;

D2. the label information of ROI macro block and each macro block carried out the decoding of Ha Fuman entropy, obtain Block _Mask, ROI _MaskAnd the quantization parameter of ROI piece;

D3., the channel estimating parameter is set, adopts the minimum rule of cross entropy to estimate channel parameter, carry out motion compensated interpolation according to the adjacent key frame in front and back of the WZ frame of having decoded out and obtain initial side information SI;

D4. utilize side information to obtain the bit stream coefficient of non-ROI piece in conjunction with the label information of each macro block by the LDPC coding;

D5. will be according to Block _MaskThe C piece that obtains, the ROI piece of entropy decoding and the non-ROI piece of block decoding are reconstructed;

2. according to claim 1 based on HVS﹠amp; The distributed coding method of ROI is characterized in that, described LDPC coding method is:

2a initializing variable node, likelihood information L (c _j) as shown in the formula:

L(c _j)＝2y _j/σ ²

Wherein, y _jFor receiving code word, σ ²Be noise variance, j=1,2 ..., n.

2b vertical direction iteration is calculated variable node likelihood value L (g _Ij), wherein variable node is the column vector of the parity check matrix H of correspondence on the bipartite graph.If h in the check matrix H _Ij=0, L (g then _Ij)=0; Otherwise be calculated as follows L (g _Ij):

L(g _ij)＝L(c _j)+∑ _i，j≠jL(h _ij)

Wherein, L (h _Ij) be the check-node likelihood information, i=1,2 ..., m, during the 1st iteration, L (h _Ij)=0.

2c horizontal direction iteration, calculation check node likelihood information L (h _Ij), wherein check-node is the row vector of the parity check matrix H of correspondence on the bipartite graph.If check matrix h _Ij=0, L (h then _Ij)=0; Otherwise be calculated as follows L (h _Ij):

$L\left(h_{\mathrm{ij}}\right)={2\tanh }^{-1}\left[{\mathrm{\&Pi;}}_{j,j\&NotEqual;i}\tanh \left(\frac{1}{2}L\left(g_{\mathrm{ij}}\right)\right)\right]$

2d deciphers judgement, obtains the discriminative information of variable node according to function obtained above, is calculated as follows variable node discriminative information L (Q _j):

L(Q _j)＝L(c _j)+∑ _iL(h _ij)

If L is (Q _jCode word is then adjudicated in)＞0

Otherwise If

Perhaps reach the maximum constraints number of times of iteration, finish, otherwise repeating step 2b.

The function that calculates in the above step is used for decoding.

3. one kind is used for implementing based on HVS﹠amp; The system of the distributive vide frequency coding method of ROI is characterized in that: this system comprises encoder;

Encoder comprises,

Described encoder comprises,

Described H.264 intra encoder: be used for the key frame of input is carried out H.264 intraframe coding, and the compressed bit stream that obtains is sent to H.264 intraframe decoder device;

Described ESI generation unit: be used for making up coarse coding side side information at coding side, Encoder-side Side Information is abbreviated as ESI, select the reference block of forward direction consecutive frame corresponding blocks FB, backward consecutive frame corresponding blocks BB and the corresponding average block AB of front and back consecutive frame and WZ frame corresponding blocks correlation maximum as ESI, input JND decision unit after each the piece computing of ESI and WZ frame; Described FB is the abbreviation of Forward Block; Described BB is the abbreviation of Backward Block; Described AB is the abbreviation of Average Block;

Described JND threshold value generation unit: the visibility JND threshold value that is used for the JND decision unit of input is generated the human visual system who is determined by background luminance and two kinds of factors of texture masking;

Described ROI threshold value generation unit: be used for the threshold value to the ROI extraction unit automatic acquisition region of interest ROI of input;

Described JND decision unit: be used for comparing JND threshold value division C macro block and AC macro block according to ESI and current 8 * 8 margin of image element; Described C macro block is the abbreviation of Copy Block; Described AC macro block is the abbreviation of Actual Coding Block;

Described gradient SAD computing unit: being used for input ROI extraction unit computing macro block gradient SAD is that the ROI extraction is prepared;

Described dct transform unit: be used for that the Wyner-Ziv frame is carried out discrete cosine transform and obtain conversion coefficient, the coefficient of same frequency position in all different transform blocks is organized the efficiency of formation band, coefficient tape is carried out Zigzag scanning ordering and is sent to quantifying unit;

Described quantifying unit: be used for the conversion coefficient that converter unit transmits is quantized to obtain quantization parameter, this quantization parameter is carried out 2 ^MThe level unified quantization; For specific coefficient tape, identical quantification symbol position is divided into one group, has formed corresponding groups of bitplanes, then is input to the LDPC encoder;

Described ROI extraction unit: be used for extracting the ROI zone according to gradient SAD and ROI threshold value, the AC macro block is divided into ROI piece and non-ROI piece;

Described LDPC encoder: be used for each bit-plane that the quantifying unit of non-ROI piece obtains and carry out independently LDPC coding;

Described entropy coding unit: be used for ROI piece, ROI _MaskAnd Block _MaskHa Fuman coding;

Described decoder comprises:

Described H.264 intraframe decoder device: be used for the key frame compressed bit stream that receives is decoded, and the key frame images of recovering is sent to the side information generation unit;

Described side information generation unit: be used for utilizing the adjacent key frame in front and back that decodes to produce side information by motion compensated interpolation, with the Block of decoding _MaskObtain the C piece; Estimate correlation between original Wyner-Ziv frame and the side information according to the motion compensated interpolation residual information, set up channel estimation model; Again side information is carried out the DCT identical with coding side, quantification and bit plane and extract, and this bit plane is sent into the LDPC decoding unit;

Described entropy decoding unit: be used for ROI piece, ROI to receiving _MaskAnd Block _MaskHa Fuman compression coding code stream decode, and with the decoding ROI _MaskBe sent to the LDPC decoder, with Block _MaskBe sent to the side information generation unit;

Described LDPC decoder: be used for according to the non-ROI piece Wyner-Ziv compressed bit stream and the side information that receive non-ROI piece being decoded, and the bit plane that decoding obtains is sent to inverse quantization unit;

Described inverse quantization unit: be used for the quantization parameter that decoding obtains the bit stream merging bit-plane of non-ROI piece and ROI piece to LDPC, according to correlation quantization parameter carried out inverse quantization and obtain conversion coefficient, and this conversion coefficient is sent to the inverse transformation unit.

Described inverse transformation unit: be used for the conversion coefficient that inverse quantization unit transmits is carried out the AC image block that inverse discrete cosine transformation is restored and sends into reconfiguration unit;

Described reconfiguration unit: for the C macro block that decoding is obtained and AC macro block reconstruct reduction WZ two field picture.

Beneficial effect:

1) uses the JND model of human visual system HVS, reduced the data of JND threshold value one lower area and processed and transmission work, reduced transmission rate, played the effect of saving energy consumption and bandwidth.

2) the present invention is directed to the dazzle extracted region ROI region of interest that motion is violent and brightness contrast is larger, and it is carried out the entropy coding, in the situation that do not increase encoder complexity as far as possible, improve subjective quality.

Do not increase in the situation of computational complexity and operand as far as possible, improve the subjective quality of downhole video, and improve the compression of video sequence ratio, reduce transmission rate, save energy consumption and bandwidth.

Description of drawings

Fig. 1 the present invention is based on HVS﹠amp; The distributed coding system architecture diagram of ROI;

Fig. 2 the present invention is based on HVS﹠amp; The distributed video coding flow chart of ROI;

Fig. 3 the present invention is based on HVS﹠amp; The distributed video decoding process figure of ROI;

Fig. 4 the present invention is based on HVS﹠amp; Distributed coding system the first experiments experiment performance map of ROI.

Fig. 5 the present invention is based on HVS﹠amp; Distributed coding system the second experiments experiment performance map of ROI.

Embodiment

Below in conjunction with accompanying drawing an enforcement of the present invention is further described:

As shown in Figure 1: of the present invention based on HVS﹠amp; The distributed coding and decoding device of ROI comprises encoder, wherein:

Described encoder:

Described H.264 intra encoder: be used for the key frame of input is carried out H.264 intraframe coding, and the compressed bit stream that obtains is sent to H.264 intraframe decoder device;

Described ESI generation unit: be used for making up coarse coding side side information at coding side, Encoder-side Side Information is abbreviated as ESI, select the reference block of forward direction consecutive frame corresponding blocks FB, backward consecutive frame corresponding blocks BB and the corresponding average block AB of front and back consecutive frame and WZ frame corresponding blocks correlation maximum as ESI, input JND decision unit after each the piece computing of ESI and WZ frame; Described FB is the abbreviation of Forward Block; Described BB is the abbreviation of Backward Block; Described AB is the abbreviation of Average Block;

Described JND threshold value generation unit: the visibility JND threshold value that is used for the JND decision unit of input is generated the human visual system who is determined by background luminance and two kinds of factors of texture masking;

Described ROI threshold value generation unit: be used for the threshold value to the ROI extraction unit automatic acquisition region of interest ROI of input;

Described JND decision unit: be used for comparing JND threshold value division C macro block and AC macro block according to ESI and current 8 * 8 margin of image element; Described C macro block is the abbreviation of Copy Block; Described AC macro block is the abbreviation of Actual Coding Block;

Described gradient SAD computing unit: being used for input ROI extraction unit computing macro block gradient SAD is that the ROI extraction is prepared;

Described dct transform unit: be used for that the Wyner-Ziv frame is carried out discrete cosine transform and obtain conversion coefficient, the coefficient of same frequency position in all different transform blocks is organized the efficiency of formation band, coefficient tape is carried out Zigzag scanning ordering and is sent to quantifying unit;

Described quantifying unit: be used for the conversion coefficient that converter unit transmits is quantized to obtain quantization parameter, this quantization parameter is carried out 2M level unified quantization; For specific coefficient tape, identical quantification symbol position is divided into one group, has formed corresponding groups of bitplanes, then is input to the LDPC encoder;

Described ROI extraction unit: be used for extracting the ROI zone according to gradient SAD and ROI threshold value, the AC macro block is divided into ROI piece and non-ROI piece;

Described LDPC encoder: be used for each bit-plane that the quantifying unit of non-ROI piece obtains and carry out independently LDPC coding;

Described entropy coding unit: be used for ROI piece, ROI _MaskAnd Block _MaskHa Fuman coding;

Described decoder comprises:

Described H.264 intraframe decoder device: be used for the key frame compressed bit stream that receives is decoded, and the key frame images of recovering is sent to the side information generation unit;

Described side information generation unit: be used for utilizing the adjacent key frame in front and back that decodes to produce side information by motion compensated interpolation, with the Block of decoding _MaskObtain the C piece; Estimate correlation between original Wyner-Ziv frame and the side information according to the motion compensated interpolation residual information, set up channel estimation model; Again side information is carried out the DCT identical with coding side, quantification and bit plane and extract, and this bit plane is sent into the LDPC decoding unit;

Described entropy decoding unit: be used for ROI piece, ROI to receiving _MaskAnd Block _MaskHa Fuman compression coding code stream decode, and with the decoding ROI _MaskBe sent to the LDPC decoder, with Block _MaskBe sent to the side information generation unit;

Described LDPC decoder: be used for according to the non-ROI piece Wyner-Ziv compressed bit stream and the side information that receive non-ROI piece being decoded, and the bit plane that decoding obtains is sent to inverse quantization unit;

Described inverse quantization unit: be used for the quantization parameter that decoding obtains the bit stream merging bit-plane of non-ROI piece and ROI piece to LDPC, according to correlation quantization parameter carried out inverse quantization and obtain conversion coefficient, and this conversion coefficient is sent to the inverse transformation unit.

Described inverse transformation unit: be used for the conversion coefficient that inverse quantization unit transmits is carried out the AC image block that inverse discrete cosine transformation is restored and sends into reconfiguration unit;

Described reconfiguration unit: for the C macro block that decoding is obtained and AC macro block reconstruct reduction WZ two field picture.

As depicted in figs. 1 and 2, based on HVS﹠amp; The encoder that the distributed decoding method of ROI uses is mainly by H.264 intra encoder, ESI generation unit, JND threshold value generation unit, ROI threshold value generation unit, JND decision unit, gradient SAD computing unit, dct transform unit, quantifying unit, ROI extraction unit, LDPC coding unit, 11 unit of entropy coding unit form.Wherein H.264 intra encoder is used for key frame is carried out intraframe coding; The ESI generation unit is used for making up coarse coding side side information at coding side; 2 unit of JND threshold value generation unit and JND decision unit are used for dividing the insensitive C macro block of human eye according to the HVS characteristic; ROI threshold value generation unit, gradient SAD computing unit, 3 unit of ROI extraction unit are used for extracting the ROI piece; Dct transform unit, quantifying unit, LDPC coding unit, 4 unit of entropy coding unit to the label information of each unit output ROI piece of front and each macro block and non-ROI piece carry out respectively the entropy coding and LDPC Wyner-Ziv encode.The concrete steps of described coding method are:

The method is made of coding method and coding/decoding method two parts; Encoder is divided into C macro block and AC macro block in conjunction with the JND model of HVS with input video Wyner-Ziv (WZ) frame, and the AC macro block is extracted ROI, and K frame and each macro block of WZ frame are carried out absolute coding; Decoder carries out combined decoding to the code stream that receives;

The concrete steps of described coding method are:

A. input video frame is divided into key frame and Wyner-Ziv frame, key frame is carried out intraframe coding, described key frame is the K frame,

H.264 intra encoder is carried out intraframe coding to key frame; The Wyner-Ziv frame is carried out HVS﹠amp; ROI Wyner-Ziv coding;

B. described HVS﹠amp; The method of ROI Wyner-Ziv coding comprises: the JND model by HVS is divided into C macro block and AC macro block with the Wyner-Ziv frame;

Described JND model method comprises:

B1. the Wyner-Ziv frame is divided into 8 * 8 piecemeals, each piecemeal is the B piece;

B2. calculate the JND threshold value of each pixel in the B piece according to following formula:

JND(x，y，t)＝f(idl(x，y，t))JND ^s(x，y)

Wherein, luminance errors value between the adjacent t moment of idl (x, y, t) expression and t-1 average frame constantly, JND ^sJND is in the spatial domain threshold value in (x, y) expression;

$\mathrm{idl}(x,y,t)=\frac{1}{2}(I(x,y,t)-I(x,y,t-1)+\stackrel{\&OverBar;}{I}(x,y,t)-\stackrel{\&OverBar;}{I}(x,y,t-1))$

JND ^s(x，y)＝T ^l(x，y)+T _Y ^t(x，y)-C _Y×Min{T ^l(x，y)，T _Y ^t(x，y)}

Wherein, T ^l(x, y) expression background luminance adapts to influence function, T _Y ^tHVS is stronger than the sensitiveness of texture compact district for level and smooth district for the reaction of (x, y) expression texture influence function, pixel corresponding coordinate in (x, y) presentation video, C _YCoefficient correlation between expression background luminance and two kinds of influencing factors of texture masking;

B3. calculate forward, backward and the mean pixel SAD of current block B piece according to following formula:

$\mathrm{SAD}=\underset{x=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{y=1}{\overset{N}{\mathrm{\&Sigma;}}}|w(x,y)-r(x,y)|$

Wherein, w (x, y) and r (x, y) represent respectively the pixel value that encoding block and reference block are located at coordinate (x, y) in the WZ frame;

B4. obtain coding side side information ESI by following formula:

ESI＝Min _SAD{FB，BB，AB}

Wherein, FB is forward direction consecutive frame corresponding blocks, and FB is the abbreviation of Forward Block; BB is backward consecutive frame corresponding blocks, and BB is the abbreviation of Backward Block; AB is the corresponding average block of front and back consecutive frame, and AB is the abbreviation of Average Block;

B5. calculate respectively consecutive frame SAD _BIn maximum

And minimum value

Note

Obtain respectively t＜SAD _BWith t＞SAD _BThe time pixel ratio ρ ₀, ρ ₁With average gradient value m ₀, m ₁, then the overall average Grad is: m=ρ ₀m ₀+ ρ ₁m ₁

B6. the variance of calculating prospect and background is:

v＝ρ ₀ρ ₁(m ₀-m ₁) ²；

B7. exist Traversal v finds out maximum v on the interval _Max, v _MaxCorresponding t is in prospect and the background variance maximum, and this moment, the t value was threshold value T _ROI

B8. according to the T that obtains _ROIValue tentatively extracts foreground image, by comparing SAD _BWith T _ROIValue come the extraction prospect, for greater than T _ROIThen the extracting section of value is carried out foreground image one time to b8 by step b5 more out as foreground image, obtains more accurate secondary T _ROI

B9. calculate the predicted distortion of each point according to the distortion function D (x, y) of ESI, obtained the predicted distortion value of each point by following formula:

D (x, y)=| ESI (x, y)-I (x, y) | wherein, ESI (x, y) and I (x, y) represent respectively ESI value and the pixel value that coordinate (x, y) is located in the WZ frame.

B10. according to following formula piece B being carried out JND differentiates, divide AC piece and C piece, become the C piece for distortion value less than or equal to the macroblock partitions of JND threshold value, and be the AC piece for distortion value greater than the macroblock partitions of JND threshold value, and obtain the two-value mask Block that this WZ two field picture piece is divided _Mask:

${\mathrm{Block}}_{\mathrm{mask}}=\left\{\begin{array}{cc}1& \left(\mathrm{\&Sigma;}{\mathrm{JND}}_{\mathrm{mask}}\right)\&GreaterEqual;\mathrm{\&epsiv;}\&CenterDot;\mathrm{Num}\\ 0& \left(\mathrm{\&Sigma;}{\mathrm{JND}}_{\mathrm{mask}}\right)<\mathrm{\&epsiv;}\&CenterDot;\mathrm{Num}\end{array}\right.$ ${\mathrm{JND}}_{\mathrm{mask}}=\left\{\begin{array}{cc}1& (\mathrm{JND}(x,y)\&GreaterEqual;D(x,y))\\ 0& (\mathrm{JND}(x,y)<D(x,y))\end{array}\right.$

Wherein, Num is the total quantity of pixel in each macro block, and ε gets 0.1.

B11. the AC piece is done DCT and 2 ^MThe level unified quantization, the extraction coefficient band also carries out Zigzag scanning, according to selected quantization matrix conversion coefficient is carried out uniform quantization; And according to the SADB of following formula calculating K frame and WZ frame same position AC piece gradient;

${\mathrm{SAD}}_B=\underset{(j,k)\&Element;B}{\mathrm{\&Sigma;}}\left|G_K\right[F(j,k)]-G_W[F(j,k)\left]\right|$

Wherein K represents that key frame, W are that Wyner-Ziv frame, B are single macro block, the pixel value that F (j, k) locates for z coordinate (j, k).

B12. adopt following formula ROI macro block decision criteria, obtain region of interest two-value mask ROI _Mask:

${\mathrm{ROI}}_{\mathrm{mask}}=\left\{\begin{array}{cc}1& ({\mathrm{SAD}}_B\&GreaterEqual;T_{\mathrm{ROI}})\\ 0& ({\mathrm{SAD}}_B<T_{\mathrm{ROI}})\end{array}\right.$

Wherein, ROI _MaskBe the two-value mask of region of interest, as macro block gradient SAD _BThen be set to 0 less than gradient T, otherwise be 1; To two-value mask ROI _MaskCarry out simple morphology and process, first corrosion and then expansion make contour of object level and smooth, extract ROI by following formula again:

$\mathrm{ROI}=\left\{\begin{array}{cc}{W}_B& (W_B\&SubsetEqual;{\mathrm{ROI}}_{\mathrm{mask}})\\ 0& (W_B\&NotSubset;{\mathrm{ROI}}_{\mathrm{mask}})\end{array}\right.$

Wherein, W _BRepresent corresponding AC macro block in the current WZ frame, if W _BCorresponding ROI _MaskBe 1 and be the ROI piece, be non-ROI piece otherwise set to 0, finally obtain the ROI region of interest of present frame;

B13. to Block _Mask, ROI _MaskAnd the ROI piece after quantizing carries out the Ha Fuman compression coding; Non-ROI piece extraction bit plane to the AC macro block carries out the LDPC coding, if the image sets of present encoding is last image sets, then withdraws from HVS﹠amp; The ROI distributed video coding;

C. the AC macro block is extracted ROI, the label information of ROI macro block and each macro block is carried out the entropy coding, non-ROI macro block is carried out encoding based on the Wyner-Ziv of LDPC (the LDPC here can quote top definition);

D. the concrete steps of described coding/decoding method are:

D1. H.264 intraframe decoder is obtained key frame;

D2. the label information of ROI macro block and each macro block carried out the decoding of Ha Fuman entropy, obtain Block _Mask, ROI _MaskAnd the quantization parameter of ROI piece;

D3., the channel estimating parameter is set, adopts the minimum rule of cross entropy to estimate channel parameter, carry out motion compensated interpolation according to the adjacent key frame in front and back of the WZ frame of having decoded out and obtain initial side information SI;

D4. utilize side information to obtain the bit stream coefficient of non-ROI piece in conjunction with the label information of each macro block by the LDPC coding;

D5. will be according to Block _MaskThe C piece that obtains, the ROI piece of entropy decoding and the non-ROI piece of block decoding are reconstructed, and wherein, need not encode at the C piece that coding side obtains, and the C piece part that intercepts corresponding side information piece in decoding end obtains finishing behind the C piece as decoded C piece.

2. according to claim 1 based on HVS﹠amp; The distributed coding method of ROI is characterized in that, described LDPC coding method is:

2a initializing variable node, likelihood information L (c _j) as shown in the formula:

L(c _j)＝2y _j/σ ²

Wherein, y _jFor receiving code word, σ ²Be noise variance, j=1,2 ..., n.

2b vertical direction iteration is calculated variable node likelihood value L (g _Ij), wherein variable node is the column vector of the parity check matrix H of correspondence on the bipartite graph.If h in the check matrix H _Ij=0, L (g then _Ij)=0; Otherwise be calculated as follows L (g _Ij):

L(g _ij)＝L(c _j)+∑ _i，j≠jL(h _ij)

Wherein, L (h _Ij) be the check-node likelihood information, i=1,2 ..., m, during the 1st iteration, L (h _Ij)=0.

2c horizontal direction iteration, calculation check node likelihood information L (h _Ij), wherein check-node is the row vector of the parity check matrix H of correspondence on the bipartite graph.If check matrix h _Ij=0, L (h then _Ij)=0; Otherwise be calculated as follows L (h _Ij):

$L\left(h_{\mathrm{ij}}\right)={2\tanh }^{-1}\left[{\mathrm{\&Pi;}}_{j,j\&NotEqual;i}\tanh \left(\frac{1}{2}L\left(g_{\mathrm{ij}}\right)\right)\right]$

2d deciphers judgement, obtains the discriminative information of variable node according to function obtained above, is calculated as follows variable node discriminative information L (Q _j):

L(Q _j)＝L(c _j)+∑ _iL(h _ij)

If L is (Q _jCode word is then adjudicated in)＞0 Otherwise

If Perhaps reach the maximum constraints number of times of iteration, finish, otherwise repeating step 2b.

The function that calculates in the above step is used for decoding.

Effect of the present invention further specifies by following experiment:

1) experiment condition

Hardware environment: CPU Intel (R) Core (TM) 2Duo CPU E7500,2.93GHz, 2.00GB internal memory;

Gop structure: the IWWWWWWW pattern, can be by 8 frames of dividing exactly as the K frame, all the other are the WZ frame; IWIWIWIW pattern, odd-numbered frame are the K frame, and even frame is the WZ frame; The SINGLE_I pattern, first frame is as the K frame in the GOP, and all the other are the WZ frame;

Reference sequences: Foreman, Carphone;

Resolution: 176 * 144;

Product is examined the sequence condition and is seen Table 1.

Table 1 cycle tests condition

The video sequence title	Foreman	Carphone
			The test frame number	300	100
Frame per second (HZ)	30	30
			Picture format	QCIF	QCIF

2) experiment content

Add up each reference sequences adopts respectively optimum prediction coding, proposes without motion-estimation encoded, JPEG coding and the present invention under above-mentioned experiment condition HVS﹠amp; The ROI distributive vide frequency coding method is encoded, and obtains the Y-PSNR PSNR of each pattern and the curve chart of code check Rata.The experimental result of each sequence is seen Fig. 4 and Fig. 5.

As seen from Figure 4, for " Foreman " sequence, and compare HVS﹠amp without motion estimation coding method; The objective Quality of recovery PSNR of ROI distributed coding method can improve 0.3～0.7dB, compares with the JPEG coding, can improve 1～3dB; " Carphone " sequence among Fig. 5 and is compared HVS﹠amp without motion estimation coding method; The objective Quality of recovery PSNR of ROI distributed coding method can improve 1～2dB, compares with the JPEG coding, can improve 3～4dB.The HVS﹠amp that the present invention proposes; ROI distributed coding system encodes, compares without motion-estimation encoded, JPEG coding method with optimum prediction, and distortion performance increases, and encoder is simple, is easy to realize.

Claims (3)

1. one kind based on HVS﹠amp; The distributive vide frequency coding method of ROI is characterized in that:

The concrete steps of described coding method are:

A. input video frame is divided into key frame and Wyner-Ziv frame, key frame (K frame) is carried out intraframe coding, described key frame is the K frame, and H.264 intra encoder is carried out intraframe coding to key frame; The Wyner-Ziv frame is carried out HVS﹠amp; ROI Wyner-Ziv coding;

B. described HVS﹠amp; The method of ROI Wyner-Ziv coding comprises: the JND model by HVS is divided into C macro block and AC macro block with the Wyner-Ziv frame;

Described JND model method comprises:

B1. the Wyner-Ziv frame is divided into 8 * 8 piecemeals, each piecemeal is the B piece;

B2. calculate the JND threshold value of each pixel in the B piece according to following formula:

JND(x，y，t)＝f(idl(x，y，t))JND ^s(x，y)

Wherein, luminance errors value between the adjacent t moment of idl (x, y, t) expression and t-1 average frame constantly, JND ^sJND is in the spatial domain threshold value in (x, y) expression;

$\mathrm{idl}(x,y,t)=\frac{1}{2}(I(x,y,t)-I(x,y,t-1)+\stackrel{\&OverBar;}{I}(x,y,t)-\stackrel{\&OverBar;}{I}(x,y,t-1))$

JND ^s(x，y)＝T ^l(x，y)+T _Y ^t(x，y)-C _Y×Min{T ^l(x，y)，T _Y ^t(x，y)}

Wherein, T ^l(x, y) expression background luminance adapts to influence function, T _Y ^tHVS is stronger than the sensitiveness of texture compact district for level and smooth district for the reaction of (x, y) expression texture influence function, pixel corresponding coordinate in (x, y) presentation video, C _YCoefficient correlation between expression background luminance and two kinds of influencing factors of texture masking;

B3. calculate forward, backward and the mean pixel SAD of current block B piece according to following formula:

$\mathrm{SAD}=\underset{x=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{y=1}{\overset{N}{\mathrm{\&Sigma;}}}|w(x,y)-r(x,y)|$

Wherein, w (x, y) and r (x, y) represent respectively the pixel value that encoding block and reference block are located at coordinate (x, y) in the WZ frame;

B4. obtain coding side side information ESI by following formula:

ESI＝Min _SAD{FB，BB，AB}

Wherein, FB is forward direction consecutive frame corresponding blocks, and FB is the abbreviation of Forward Block; BB is backward consecutive frame corresponding blocks, and BB is the abbreviation of Backward Block; AB is the corresponding average block of front and back consecutive frame, and AB is the abbreviation of Average Block;

B5. calculate respectively consecutive frame SAD _BIn maximum

And minimum value

Note Obtain respectively t＜SAD _BWith t＞SAD _BThe time pixel ratio ρ ₀, ρ ₁With average gradient value m ₀, m ₁, then the overall average Grad is: m=ρ ₀m ₀+ ρ ₁m ₁

B6. the variance of calculating prospect and background is:

v＝ρ ₀ρ ₁(m ₀-m ₁) ²；

B7. exist

Traversal v finds out maximum v on the interval _Max, v _MaxCorresponding t is in prospect and the background variance maximum, and this moment, the t value was threshold value T _ROI

B8. according to the T that obtains _ROIValue tentatively extracts foreground image, by comparing SAD _BWith T _ROIValue extract foreground image, for greater than T _ROIThen the extracting section of value is carried out foreground image one time to b8 by step b5 more out as foreground image, obtains more accurate secondary T _ROI

B9. calculate the predicted distortion of each point according to the distortion function D (x, y) of ESI, obtained the predicted distortion value of each point by following formula:

D (x, y)=| ESI (x, y)-I (x, y) | wherein, ESI (x, y) and I (x, y) represent respectively ESI value and the pixel value that coordinate (x, y) is located in the WZ frame.

B10. according to following formula piece B being carried out JND differentiates, divide AC piece and C piece, become the C piece for distortion value less than or equal to the macroblock partitions of JND threshold value, and be the AC piece for distortion value greater than the macroblock partitions of JND threshold value, and obtain the two-value mask Block that this WZ two field picture piece is divided _Mask:

${\mathrm{Block}}_{\mathrm{mask}}=\left\{\begin{array}{cc}1& \left(\mathrm{\&Sigma;}{\mathrm{JND}}_{\mathrm{mask}}\right)\&GreaterEqual;\mathrm{\&epsiv;}\&CenterDot;\mathrm{Num}\\ 0& \left(\mathrm{\&Sigma;}{\mathrm{JND}}_{\mathrm{mask}}\right)<\mathrm{\&epsiv;}\&CenterDot;\mathrm{Num}\end{array}\right.$ ${\mathrm{JND}}_{\mathrm{mask}}=\left\{\begin{array}{cc}1& (\mathrm{JND}(x,y)\&GreaterEqual;D(x,y))\\ 0& (\mathrm{JND}(x,y)<D(x,y))\end{array}\right.$

Wherein, Num is the total quantity of pixel in each macro block, and ε gets 0.1.

B11. the AC piece is done DCT and 2 ^MThe level unified quantization, the extraction coefficient band also carries out Zigzag scanning, according to selected quantization matrix conversion coefficient is carried out uniform quantization; And according to the SAD of following formula calculating K frame and WZ frame same position AC piece gradient _B

${\mathrm{SAD}}_B=\underset{(j,k)\&Element;B}{\mathrm{\&Sigma;}}\left|G_K\right[F(j,k)]-G_W[F(j,k)\left]\right|$

Wherein K represents that key frame, W are that Wyner-Ziv frame, B are single macro block, the pixel value that F (j, k) locates for z coordinate (j, k).

B12. adopt following formula ROI macro block decision criteria, obtain region of interest two-value mask ROI _Mask:

${\mathrm{ROI}}_{\mathrm{mask}}=\left\{\begin{array}{cc}1& ({\mathrm{SAD}}_B\&GreaterEqual;T_{\mathrm{ROI}})\\ 0& ({\mathrm{SAD}}_B<T_{\mathrm{ROI}})\end{array}\right.$

Wherein, ROI _MaskBe the two-value mask of region of interest, as macro block gradient SAD _BThen be set to 0 less than gradient T, otherwise be 1; To two-value mask ROI _MaskCarry out simple morphology and process, first corrosion and then expansion make contour of object level and smooth, extract ROI by following formula again:

$\mathrm{ROI}=\left\{\begin{array}{cc}{W}_B& (W_B\&SubsetEqual;{\mathrm{ROI}}_{\mathrm{mask}})\\ 0& (W_B\&NotSubset;{\mathrm{ROI}}_{\mathrm{mask}})\end{array}\right.$

Wherein, W _BRepresent corresponding AC macro block in the current WZ frame, if W _BCorresponding ROI _MaskBe 1 and be the ROI piece, be non-ROI piece otherwise set to 0, finally obtain the ROI region of interest of present frame;

B13. to Block _Mask, ROI _MaskAnd the ROI piece after quantizing carries out the Ha Fuman compression coding; Non-ROI piece extraction bit plane to the AC macro block carries out the LDPC coding, if the image sets of present encoding is last image sets, then withdraws from HVS﹠amp; The ROI distributed video coding;

C. the AC macro block is extracted ROI, the label information of ROI macro block and each macro block is carried out the entropy coding, non-ROI macro block is carried out encoding based on the Wyner-Ziv of LDPC (the LDPC here can quote top definition);

D. the concrete steps of described coding/decoding method are:

D1. H.264 intraframe decoder is obtained key frame;

D2. the label information of ROI macro block and each macro block carried out the decoding of Ha Fuman entropy, obtain Block _Mask, ROI _MaskAnd the quantization parameter of ROI piece;

D3., the channel estimating parameter is set, adopts the minimum rule of cross entropy to estimate channel parameter, carry out motion compensated interpolation according to the adjacent key frame in front and back of the WZ frame of having decoded out and obtain initial side information SI;

D4. utilize side information to obtain the bit stream coefficient of non-ROI piece in conjunction with the label information of each macro block by the LDPC coding;

D5. will be according to Block _MaskThe C piece that obtains, the ROI piece of entropy decoding and the non-ROI piece of block decoding are reconstructed;

2. according to claim 1 and 2 based on HVS﹠amp; The distributed coding method of ROI is characterized in that, described LDPC coding method is:

2a initializing variable node, likelihood information L (c _j) as shown in the formula:

L(c _j)＝2y _j/σ ²

Wherein, y _jFor receiving code word, σ ²Be noise variance, j=1,2 ..., n.

2b vertical direction iteration is calculated variable node likelihood value L (g _Ij), wherein variable node is the column vector of the parity check matrix H of correspondence on the bipartite graph.If h in the check matrix H _Ij=0, L (g then _Ij)=0; Otherwise be calculated as follows L (g _Ij):

L(g _ij)＝L(c _j)+∑ _i，j≠jL(h _ij)

Wherein, L (h _Ij) be the check-node likelihood information, i=1,2 ..., m, during the 1st iteration, L (h _Ij)=0.

2c horizontal direction iteration, calculation check node likelihood information L (h _Ij), wherein check-node is the row vector of the parity check matrix H of correspondence on the bipartite graph.If check matrix h _Ij=0, L (h then _Ij)=0; Otherwise be calculated as follows L (h _Ij):

$L\left(h_{\mathrm{ij}}\right)={2\tanh }^{-1}\left[{\mathrm{\&Pi;}}_{j,j\&NotEqual;i}\tanh \left(\frac{1}{2}L\left(g_{\mathrm{ij}}\right)\right)\right]$

2d deciphers judgement, obtains the discriminative information of variable node according to function obtained above, is calculated as follows variable node discriminative information L (Q _j):

L(Q _j)＝L(c _j)+∑ _iL(h _ij)

If L is (Q _jCode word is then adjudicated in)＞0 Otherwise

If Perhaps reach the maximum constraints number of times of iteration, finish, otherwise repeating step 2b.

The function that calculates in the above step is used for decoding.

3. one kind is used for implementing based on HVS﹠amp; The system of the distributive vide frequency coding method of ROI is characterized in that: this system comprises encoder;

Encoder comprises,

Described encoder comprises,

Described H.264 intra encoder: be used for the key frame of input is carried out H.264 intraframe coding, and the compressed bit stream that obtains is sent to H.264 intraframe decoder device;

Described ESI generation unit: be used for making up coarse coding side side information at coding side, Encoder-side Side Information is abbreviated as ESI, select the reference block of forward direction consecutive frame corresponding blocks FB, backward consecutive frame corresponding blocks BB and the corresponding average block AB of front and back consecutive frame and WZ frame corresponding blocks correlation maximum as ESI, input JND decision unit after each the piece computing of ESI and WZ frame; Described FB is the abbreviation of Forward Block; Described BB is the abbreviation of Backward Block; Described AB is the abbreviation of Average Block;

Described JND threshold value generation unit: the visibility JND threshold value that is used for the JND decision unit of input is generated the human visual system who is determined by background luminance and two kinds of factors of texture masking;

Described ROI threshold value generation unit: be used for the threshold value to the ROI extraction unit automatic acquisition region of interest ROI of input;

Described JND decision unit: be used for comparing JND threshold value division C macro block and AC macro block according to ESI and current 8 * 8 margin of image element; Described C macro block is the abbreviation of Copy Block; Described AC macro block is the abbreviation of Actual Coding Block;

Described gradient SAD computing unit: being used for input ROI extraction unit computing macro block gradient SAD is that the ROI extraction is prepared;

Described dct transform unit: be used for that the Wyner-Ziv frame is carried out discrete cosine transform and obtain conversion coefficient, the coefficient of same frequency position in all different transform blocks is organized the efficiency of formation band, coefficient tape is carried out Zigzag scanning ordering and is sent to quantifying unit;

Described quantifying unit: be used for the conversion coefficient that converter unit transmits is quantized to obtain quantization parameter, this quantization parameter is carried out 2 ^MThe level unified quantization; For specific coefficient tape, identical quantification symbol position is divided into one group, has formed corresponding groups of bitplanes, then is input to the LDPC encoder;

Described ROI extraction unit: be used for extracting the ROI zone according to gradient SAD and ROI threshold value, the AC macro block is divided into ROI piece and non-ROI piece;

Described LDPC encoder: be used for each bit-plane that the quantifying unit of non-ROI piece obtains and carry out independently LDPC coding;

Described entropy coding unit: be used for ROI piece, ROI _MaskAnd Block _MaskHa Fuman coding;

Described decoder comprises:

Described H.264 intraframe decoder device: be used for the key frame compressed bit stream that receives is decoded, and the key frame images of recovering is sent to the side information generation unit;

Described side information generation unit: be used for utilizing the adjacent key frame in front and back that decodes to produce side information by motion compensated interpolation, with the Block of decoding _MaskObtain the C piece; Estimate correlation between original Wyner-Ziv frame and the side information according to the motion compensated interpolation residual information, set up channel estimation model; Again side information is carried out the DCT identical with coding side, quantification and bit plane and extract, and this bit plane is sent into the LDPC decoding unit;

Described entropy decoding unit: be used for ROI piece, ROI to receiving _MaskAnd Block _MaskHa Fuman compression coding code stream decode, and with the decoding ROI _MaskBe sent to the LDPC decoder, with Block _MaskBe sent to the side information generation unit;

Described LDPC decoder: be used for according to the non-ROI piece Wyner-Ziv compressed bit stream and the side information that receive non-ROI piece being decoded, and the bit plane that decoding obtains is sent to inverse quantization unit;

Described inverse quantization unit: be used for the quantization parameter that decoding obtains the bit stream merging bit-plane of non-ROI piece and ROI piece to LDPC, according to correlation quantization parameter carried out inverse quantization and obtain conversion coefficient, and this conversion coefficient is sent to the inverse transformation unit.

Described inverse transformation unit: be used for the conversion coefficient that inverse quantization unit transmits is carried out the AC image block that inverse discrete cosine transformation is restored and sends into reconfiguration unit;

Described reconfiguration unit: for the C macro block that decoding is obtained and AC macro block reconstruct reduction WZ two field picture.

Images