CN101917627B - Video fault-tolerant coding method based on self-adaptation flexible macro-block order - Google Patents

Abstract

The invention discloses a video fault-tolerant coding method based on a self-adaptation flexible macro-block order, which comprises the steps of: inputting a frame of image and encoding; extracting a motion area of the image, estimating fault sensitivity of the motion area, and deciding an optimum flexible macro-block order coding mode; calculating the accumulated interframe differences of all macro-block positions, compulsorily carrying out intraframe refreshing on the macro block positions at the next frame when the accumulated interframe differences in the macro-block position in the motion area is larger than an appointed refreshing threshold, and accordingly, calculating the quantity and positions of the refreshed intraframe macro blocks of the motion area; forming a coding scheme of the next frame of image according to the flexible macro-block order coding mode and a refreshed macro-block calculation result; and repeating the steps until all the coding are completed. The video fault-tolerant coding method not only can meet the requirement of differentiated network transmission, but also can fully exert the fault-tolerant function of FMO (Flexible Macro-Block Order), thereby realizing better fault-tolerance effect with less computing resource consumption.

Description

A kind of video fault-tolerant coding method based on self-adaptation flexible macro-block order

Technical field

The present invention relates to the coding and decoding video field, be specifically related to a kind of video fault-tolerant coding method based on self-adaptation flexible macro-block order (FMO).

Background technology

The fast development of the Along with computer technology and the network communications technology, simple text message form can not satisfy people's demand, and more rich video information has been come into the various aspects of people's daily life gradually.Limited transmission bandwidth and memory space inevitable requirement are compressed the huge original video of data volume, so video coding and decoding technology and standard have become part indispensable in the multimedia application.Existing video encoding and decoding standard depends on the integrality of code stream very much.In case the viewing quality decline after mistake will cause video decode in transmission course, occurs, influence user's impression.Therefore, in video coding process, strengthening code stream and resist wrong ability, promptly improve the fault-tolerance of code stream, is very necessary.

Flexible macro-block order FMO (Flexible Macroblock Ordering) is a kind of very effective new fault tolerant mechanism adopted of standard H.264/AVC.Macro block no longer must be that strictness is dispensed in the sheet according to its raster scan order in image.(Macroblock allocation map, MBAmap), a macro block can be dispensed in any one sheet group and go through macroblock allocation figure.The reason that FMO mechanism can be used in the video error tolerance mainly contains two aspects.First aspect is its network environment that is very suitable for existing differential transmission protection.Owing to can carry out the sheet set of dispense of arbitrary form to macro block, therefore important or fragile zone just can be dispensed on transmission with respect in the reliable sheet group, has strengthened the specific aim of protection, has improved the efficient of protection.On the other hand, FMO itself also possesses stronger fault tolerance.Because the macro block of losing may be surrounded by its other macro block of not losing of periphery, therefore utilize the pixel of peripheral macro block or the macro block that information can have efficient recovery to lose, very help the error resilience algorithms of decoder.

Though the existing method of utilizing FMO to carry out fault-tolerant encoding has obtained certain effect, but still the problem that exists mainly comprises: do not consider transmission characteristic; Need carry out encoding the second time or adopting the rate-distortion optimization algorithm, amount of calculation and structure are not suitable for practical application; Can only in the off-line coding, carry out the non real-time operation; When paying close attention to differential transmission, ignored the fault tolerance of performance FMO itself, perhaps opposite.

Summary of the invention

The invention provides a kind of video fault-tolerant coding method, can either satisfy the Network Transmission requirement of differentiation, can give full play to the fault tolerance of FMO itself again, realize that effective video is fault-tolerant based on self-adaptation flexible macro-block order (FMO); Simultaneously, because main computing information comes from encoding domain, the information that promptly produces in the cataloged procedure, so computational efficiency is higher.

A kind of video fault-tolerant coding method based on self-adaptation flexible macro-block order (FMO) may further comprise the steps:

(1) input one two field picture and coding;

(2) moving region of the described image of extraction, the error sensitivity of estimation moving region; According to the error sensitivity of moving region, the flexible macro-block order coding mode that decision-making is optimum;

(3) calculate the accumulation frame difference of all macro block position; When the accumulation frame difference of macro block position in the moving region greater than appointment refresh threshold value the time; Described macro block position is forced frame refreshing when next frame, calculate the quantity and the position of the frame refreshing macro block of moving region thus;

The quantity and the position of the frame refreshing macro block that flexible macro-block order coding mode that (4) obtains according to step (2) and step (3) obtain form the encoding scheme of next frame image;

(5) repeating step (1)～(4) are all over until coding.

In the step (2), the moving region of the described image of described extraction comprises:

(2.1) the frame difference information of utilizing motion search to produce is carried out initial motion to macro block and is detected, and the difference between incoming frame and the reference frame is regarded as the initial motion macro block greater than the macro block of motion detection threshold;

Coordinate is (x, the initial motion sign M of macro block y) in the n two field picture _InitialShown in (I):

$M_{\mathrm{initial}}(x,y,n)=\left\{\begin{array}{cc}1,& \mathrm{if}{\mathrm{sad}}_{\mathrm{mv}00}(x,y,n)>[\mathrm{\β}\left(n\right)\×{\mathrm{sad}}_{\mathrm{avg}}\left(n\right)]\\ 0,& \mathrm{otherwise}\end{array}\right.---\left(I\right)$

In the formula (I), sad _Mv00The SAD that obtains when being 16 * 16 motion search (Sum of Absolute Difference) coded message at the zero motion vector place, the difference of reflection macro block between incoming frame and reference frame; Sad _AvgRepresent all macro block sad of n two field picture _Mv00Mean value; β (n) is the motion detection threshold parameter, [β (n) * sad _Avg(n)] be motion detection threshold.

Sad when macro block _Mv00During greater than motion detection threshold, initial motion sign M _InitialBe 1, macro block is regarded as the initial motion macro block; Otherwise, initial motion sign M _InitialBe 0, macro block is regarded as non-initial motion macro block.

Further, described motion detection threshold is along with the characteristic of not homotactic characteristic and same sequence different frame is dynamically adjusted, and wherein, motion detection threshold parameter beta (n) is suc as formula shown in (II):

In the formula (II), σ _FrameRepresent all macro block sad of n two field picture _Mv00Standard deviation; K, σ ₀, β _H, β _L, σ _{MA_H}And σ _{MA_L}Be the fixed value parameter;

(2.2) analyzes the macroblock encoding bit value, described initial motion macro block is carried out filtering, confirm the motion macro block, and the rectangle of confirming to contain the minimum of all motion macro blocks is the moving region:

Calculate the quantity N of initial motion macro block in four peripheral macro blocks of each initial motion macro block _Neb,

For N _NebValue is 0 or 1 initial motion macro block,

Its final motion sign M _FinalFor:

$M_{\mathrm{final}}(x,y,n)=\left\{\begin{array}{cc}1,& \mathrm{if}{B}_{\mathrm{mb}}(x,y,n)\≥[{\mathrm{\λ}}_{\mathrm{filter}}\×B_{\mathrm{avg}}\left(n\right)]\\ 0,& \mathrm{otherwise}\end{array}\right.---\left(\mathrm{III}\right)$

In the formula (III), B _MbRepresent a macroblock encoding bit value, B _AvgRepresent all N _NebThe mean value of initial motion macroblock coding bit value greater than 1; λ _FilterFor the fixed value parameter, for N _NebBe 0 and 1 initial motion macro block, λ _FilterBe made as 1.4 and 0.7 respectively.

When the macroblock encoding bit value more than or equal to filtering threshold [λ _Filter* B _Avg(n)] time, final motion sign M _FinalBe 1, be the motion macro block; Otherwise, final motion sign M _FinalBe 0, be non-motion macro block.

For non-initial motion macro block and N _NebValue is greater than other initial motion macro block of 1,

M _final＝M _initial

Thus, obtain the M of all macro blocks _FinalValue, all final motion sign M _FinalValue is that 1 macro block is the motion macro block, and the rectangle of containing the minimum of all motion macro blocks is the moving region.

In the step (2), the error sensitivity of described moving region is meant that the moving region carries out the wrong difficulty of recovering in decoder, adopts and estimates as shown in the formula the calculating formula shown in (IV):

S _MA(n)＝p(n)×T(n)×σ _mb(n)×σ _frame(n) (IV)

In the formula (IV), p is the transmission packet loss of code stream, and more serious transmission packet loss can cause bigger mistake to recover difficulty; T is the texture complexity of moving region, and complicated texture can add the difficulty that serious mistake recovers; σ _MbBe macro block movement differential in the moving region, the motion conditions difference between macro block is big more, and the difficulty that mistake is recovered also can be big more; σ _FrameBe regional movement intensity, the overall movement intensity of moving region is big more with respect to background, and the difficulty that mistake is recovered also can bigger (regional movement intensity σ _FrameWith σ in the formula (II) _FrameIdentical, can utilize before all macro block sad of n two field picture _Mv00The result of calculation of standard deviation).Above-mentioned analytical factor can adapt to the algorithm characteristics of most of error recovery methods.The result is big more in the formula (IV), and the error sensitivity in account for motion zone is high more, and promptly wrong recovery difficulty is big more.

The texture complexity T of described moving region is the texture complexity t of all macro blocks in the moving region _MbMean value, and the texture complexity t of a macro block _MbCalculating is as shown in the formula shown in (V):

$t_{\mathrm{mb}}(x,y,n)=\frac{\underset{i,j\∈\mathrm{\Ω}(x,y,n)}{\mathrm{\Σ}}{\mathrm{sad}}_{\mathrm{DC}44}(i,j,x,y,n)}{N_{\mathrm{\Ω}}(x,y,n)}---\left(V\right)$

In the formula (V), Ω representes 4 * 4 the set that the neighboring pixel of the left side and top is all complete, N _ΩRepresent 4 * 4 quantity in this set, coordinate (i, j) one 4 * 4 positions in its place macro block of expression, sad _DC44The SAD information of the intra-frame 4 * 4 DC mode that produces during for mode decision; It is the difference sum of 4 * 4 all composition pixels and neighboring pixel (pixel of the left side and top) average; Reflect the otherness between these 4 * 4 each pixels of inside, difference large texture more is more complicated.

In the step (2), described error sensitivity according to the moving region, the process of the flexible macro-block order coding mode that decision-making is optimum is following:

The monolithic group coding is all used in non-moving region;

For the moving region, the error sensitivity S when the moving region _MAGreater than the susceptibility threshold S that sets _THThe time, adopt the checker board pattern of two sheet groups to encode; Error sensitivity S when the moving region _MALess than the susceptibility threshold S that sets _THThe time, the smooth pattern-coding of a sheet group of employing.

In the step (3), the accumulation frame difference of described macro block position is obtained by following formula (VI):

$D_{\mathrm{acc}}(x,y,n)=\underset{t=n_0}{\overset{n}{\mathrm{\Σ}}}p\left(t\right)\×{\mathrm{sad}}_{\mathrm{mv}00}(x,y,t)---\left(\mathrm{VI}\right)$

In the formula (VI), n ₀Expression macro block position (x, the image sequence when y) last time carrying out intraframe predictive coding number, n ₀Be less than or equal to n; P is the transmission packet loss of code stream;

In the step (3), described appointment refresh threshold value D _TH(n) be variate, calculate by following formula (VII);

$D_{\mathrm{TH}}\left(n\right)=\frac{C_{\mathrm{IR}}}{{\mathrm{\σ}}_{\mathrm{frame}}\left(n\right)}---\left(\mathrm{VII}\right)$

In the formula (VII), C _IRBe fixed value parameter, σ _FrameBe regional movement intensity.Give more to refresh quantity for the big frame of exercise intensity, to obtain better adaptivity.

Among the present invention, not adopting FMO to encode for first two field picture, since second frame, all is the fault-tolerant encoding patterns that decide next frame through the analysis to former frame encoding domain information.Main computing information all comes from encoding domain, and the information that is just produced in the cataloged procedure has been avoided repeatedly coding, has effectively saved computational resource.

Among the present invention, when extracting the moving region, when taking 16 * 16 motion search at the coded message SAD at zero motion vector place (sad _Mv00) carry out preliminary motion detection, because sad _Mv00Can reflect the difference between incoming frame and the reference frame and can directly from most of searching algorithms, obtain, therefore can detect the moving region expeditiously; After having carried out preliminary motion detection, also adopt the macroblock coding bit value that the initial motion macro block is carried out filtering, reject the noise macro block in the initial motion macro block, remove remaining noise.Because the noise macro block often is in isolated positions, therefore,, confirm the noise macro block through analyzing the encoding ratio paricular value size of isolated macroblock encoding bit value with respect to the primary motor area territory.Initial motion macro block behind the cancelling noise macro block is confirmed as the motion macro block.The present invention is to be the basis with the video monitoring scene, and the moving region that it is most crucial (no matter being people or object) as the object of prioritised transmission and focused protection, and the rectangle of confirming to contain the minimum of all motion macro blocks is the moving region.

Among the present invention, according to macro block movement differential σ in the texture complexity T of the transmission packet loss p of code stream, moving region, the moving region _MbWith regional movement intensity σ _FrameEtc. the error sensitivity of a plurality of factor assessments moving region, can adapt to the algorithm characteristics of most of error recovery methods.In addition, when the obtaining of texture information, the SAD information (sad of the intra-frame 4 * 4 DC mode that produces when utilizing mode decision _DC44), improved computational efficiency; And, macro block movement differential σ in the texture complexity T of moving region and the moving region _MbCalculating all be limited in the moving region, reduced the complexity of calculating.

Among the present invention, the selection of optimum FMO coding mode fully takes into account the difference of moving region of difference and the different error sensitivities of motion and non-moving region.In various standard FMO types, the type 2 of two groups is fit to differentiation motion (prospect) and non-moving region (background) one type beyond doubt very much, is particularly suitable for the differential transmission protection to the image zones of different.Yet type 2 protection to the moving region in some cases is not enough, because it has only adopted a kind of coded system of smooth formula, and monolithic group coding moving region just.But the type has compression efficiency preferably; On the contrary, when adopting the Class1 of checkerboard type, fault-tolerant ability can significantly strengthen.But because the information share between the adjacent macroblocks significantly reduces, the compression efficiency of Class1 is relatively poor.Therefore, the present invention has adopted and has all used the monolithic group to improve compression efficiency to background.This is because the background area is not the key area of paying close attention on the one hand, be on the other hand because few moving background comparatively speaking and be not easy to receive the influence of error of transmission.And in order between compression efficiency and fault-tolerant ability, to average out, according to the error sensitivity of prospect, with one or two sheet group prospect is encoded.FMO mechanism can provide a close friend's macro block distributed architecture to decoder among the present invention, therefore can alleviate mistake and spread.But FMO mechanism itself can not stop this mistake fully and spread, and therefore, the present invention further utilizes intra-frame macro block updating mechanism to go to strengthen its fault-tolerance, position that reasonable computation refreshes and quantity.

In sum, the video fault-tolerant coding method based on self-adaptation flexible macro-block order (FMO) of the present invention can either satisfy the Network Transmission requirement of differentiation, can give full play to the fault tolerance of FMO itself again, realizes that effective video is fault-tolerant; Simultaneously, because main computing information comes from encoding domain, the information that promptly produces in the cataloged procedure, so computational efficiency is higher.

Description of drawings

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

Fig. 2 is the schematic flow sheet of the moving region of extraction image;

Fig. 3 is a self adaptation FMO coding mode decision-making sketch map;

Fig. 4 is the PSNR result of Salesman sequence;

Fig. 5 is the PSNR result of Container sequence;

Fig. 6 is the PSNR result of Hall sequence.

Embodiment

Specify the present invention below in conjunction with embodiment and accompanying drawing, but the present invention is not limited to this.

As shown in Figure 1, a kind of video fault-tolerant coding method based on self-adaptation flexible macro-block order (FMO) may further comprise the steps:

(1) input one two field picture and coding;

Do not adopt FMO to encode for first two field picture, but adopt the normal encoding mode.

(2) moving region of the described image of extraction, the error sensitivity of estimation moving region; And according to the error sensitivity of moving region, the flexible macro-block order coding mode that decision-making is optimum;

Wherein, the extraction of the moving region of described image comprises:

(2.1) the frame difference information of utilizing motion search to produce is carried out initial motion to macro block and is detected, and the difference between incoming frame and the reference frame is regarded as the initial motion macro block greater than the macro block of motion detection threshold; Coordinate is (x, the initial motion sign M of macro block y) in the n two field picture _InitialShown in (I):

$M_{\mathrm{initial}}(x,y,n)=\left\{\begin{array}{cc}1,& \mathrm{if}{\mathrm{sad}}_{\mathrm{mv}00}(x,y,n)>[\mathrm{\β}\left(n\right)\×{\mathrm{sad}}_{\mathrm{avg}}\left(n\right)]\\ 0,& \mathrm{otherwise}\end{array}\right.---\left(I\right)$

In the formula (I), sad _Mv00In the coded message at zero motion vector place, the difference between reflection incoming frame and the reference frame can directly obtain from most of searching algorithms when being 16 * 16 motion search; Sad _AvgRepresent all macro block sad of n two field picture _Mv00Mean value; β (n) is the motion detection threshold parameter, [β (n) * sad _Avg(n)] be motion detection threshold.

Sad when macro block _Mv00During greater than motion detection threshold, initial motion sign M _InitialBe 1, macro block is regarded as the initial motion macro block; Otherwise, initial motion sign M _InitialBe 0, macro block is regarded as non-initial motion macro block.

Further, described motion detection threshold is along with the characteristic of not homotactic characteristic and same sequence different frame is dynamically adjusted, and wherein, motion detection threshold parameter beta (n) is suc as formula shown in (II):

In the formula (II), σ _FrameRepresent all macro block sad of n two field picture _Mv00Standard deviation; K, σ ₀, β _H, β _L, σ _{MA_H}And σ _{MA_L}Be the fixed value parameter;

(2.2) after having carried out preliminary motion detection; Also to carry out filtering, some initial motion macro blocks rejected away, therefore the noise of remnants; Utilize the macroblock encoding bit value; The initial motion macro block is carried out filtering, confirm the motion macro block, and the rectangle of confirming to contain the minimum of all motion macro blocks is the moving region:

The noise macro block often is in isolated positions, therefore, calculates the quantity N of initial motion macro block in four peripheral macro blocks of each initial motion macro block earlier _Neb, find the initial motion macro block that is in isolated positions to be isolated macro block, also promptly: N _NebValue is 0 or 1 initial motion macro block; Further, if isolated macro block is not enough enough big with respect to the encoding ratio paricular value in primary motor area territory, isolated macro block just will be identified as the noise macro block.That is:

For N _NebValue is 0 or 1 initial motion macro block,

Final motion sign M _FinalFor:

$M_{\mathrm{final}}(x,y,n)=\left\{\begin{array}{cc}1,& \mathrm{if}{B}_{\mathrm{mb}}(x,y,n)\≥[{\mathrm{\λ}}_{\mathrm{filter}}\×B_{\mathrm{avg}}\left(n\right)]\\ 0,& \mathrm{otherwise}\end{array}\right.---\left(\mathrm{III}\right)$

In the formula (III), B _MbRepresent a macroblock encoding bit value, B _AvgRepresent all N _NebThe mean value of initial motion macroblock coding bit value greater than 1; λ _FilterFor the fixed value parameter, for N _NebBe 0 and 1 initial motion macro block, λ _FilterBe made as 1.4 and 0.7 respectively.

When the macroblock encoding bit value greater than filtering threshold [λ _Filter* B _Avg(n)] time, final motion sign M _FinalBe 1, be the motion macro block; Otherwise, final motion sign M _FinalBe 0, be noise macro block (non-motion macro block).

For non-initial motion macro block and N _NebValue is greater than other initial motion macro block of 1,

M _final＝M _initial

Thus, obtain the M of all macro blocks _FinalValue, all final motion sign M _FinalValue is that 1 macro block is the motion macro block, and the rectangle of containing the minimum of all motion macro blocks is defined as the moving region.

The all processes that extract for the moving region above-mentioned steps (2.1) and (2.2), as shown in Figure 2.In the present embodiment; With the video monitoring scene is the basis; The moving region that it is most crucial (no matter being people or object) is as the object of prioritised transmission and focused protection; Finally have only a moving region to be extracted out and as prospect, all other macro blocks belong to background, promptly moving region and prospect have identical implication.

In the step (2), the error sensitivity of described moving region is meant that the moving region carries out the wrong difficulty of recovering in decoder, adopts and estimates as shown in the formula the calculating formula shown in (IV):

S _MA(n)＝p(n)×T(n)×σ _mb(n)×σ _frame(n) (IV)

In the formula (IV), p is the transmission packet loss of code stream, and more serious transmission packet loss can cause bigger mistake to recover difficulty; T is the texture complexity of moving region, and complicated texture can add the difficulty that serious mistake recovers; σ _MbBe macro block movement differential in the moving region, the motion conditions difference between macro block is big more, and the difficulty that mistake is recovered also can be big more; σ _FrameBe regional movement intensity, the overall movement intensity of moving region is big more with respect to background, and the difficulty that mistake is recovered also can bigger (regional movement intensity σ _FrameWith σ in the formula (II) _FrameIdentical, can utilize before all macro block sad of n two field picture _Mv00The result of calculation of standard deviation).Above-mentioned analytical factor can adapt to the algorithm characteristics of most of error recovery methods.The result is big more in the formula (IV), and the error sensitivity in account for motion zone is high more, and promptly wrong recovery difficulty is big more.

The texture complexity T of described moving region is the texture complexity t of all macro blocks in the moving region _MbMean value, and the texture complexity t of a macro block _MbCalculating is as shown in the formula shown in (V):

$t_{\mathrm{mb}}(x,y,n)=\frac{\underset{i,j\∈\mathrm{\Ω}(x,y,n)}{\mathrm{\Σ}}{\mathrm{sad}}_{\mathrm{DC}44}(i,j,x,y,n)}{N_{\mathrm{\Ω}}(x,y,n)}---\left(V\right)$

In the formula (V), Ω representes the left side and 4 * 4 complete set of top pixel, does not consider the unavailable minority divided block of the left side or top pixel, as is in the piece on the border of image or sheet; N _ΩRepresent 4 * 4 quantity in this set, coordinate (i, j) one 4 * 4 positions in its place macro block of expression, sad _DC44The SAD information of the intra-frame 4 * 4 DC mode that produces during for mode decision; It is the difference sum of 4 * 4 all composition pixels and neighboring pixel (left side and top) average; Reflect the otherness between these 4 * 4 each pixels of inside, difference large texture more is more complicated.Be similar to σ _Frame, σ _MbBe a standard deviation, but its computer capacity is limited to moving region rather than entire image.T and σ _MbThe calculating of the two all is to be limited in the moving region, to reduce the complexity of calculating.

In the step (2), described error sensitivity according to the moving region, the process of the flexible macro-block order coding mode that decision-making is optimum is following:

All use the monolithic group coding to improve compression efficiency to non-moving region (background), because the background area is not the key area of paying close attention to, and few moving background comparatively speaking and be not easy to receive the influence of error of transmission;

For the moving region, according to the error sensitivity of moving region (prospect), prospect is encoded, so that between compression efficiency and fault-tolerant ability, average out with one or two sheet group.The concrete realization is the basis with the type 6 of FMO.Error sensitivity S when moving region (prospect) _MAGreater than the susceptibility threshold S that sets _THThe time, adopt the checker board pattern of two sheet groups to encode; Error sensitivity S when the moving region _MALess than the susceptibility threshold S that sets _THThe time, the smooth pattern-coding of a sheet group of employing, this moment is identical with the structure of the type 2 of two groups.These two kinds of FMO coding modes are as shown in Figure 3.Arabic numerals among the figure are represented the sheet group number under each macro block, and the border of sheet group is represented with thick lines, the inner moving region for extracting of thick lines.

(3) calculate the accumulation frame difference of all macro block position; When the accumulation frame difference of macro block position in the moving region greater than appointment refresh threshold value the time; Described macro block position is forced frame refreshing when next frame, calculate the quantity and the position of the frame refreshing macro block of moving region thus;

For significantly not increasing amount of calculation, aforesaid sad _Mv00Information is used continuing at this; Simultaneously, also need consider the factor of packet loss, because serious packet loss can cause spreading than grave error.The accumulation frame difference of described macro block position is obtained by following formula (VI):

$D_{\mathrm{acc}}(x,y,n)=\underset{t=n_0}{\overset{n}{\mathrm{\Σ}}}p\left(t\right)\×{\mathrm{sad}}_{\mathrm{mv}00}(x,y,t)---\left(\mathrm{VI}\right)$

In the formula (VI), described appointment refresh threshold value D _TH(n) be variate, by following n ₀Expression macro block position (x, the image sequence when y) last time carrying out intraframe predictive coding number, n ₀Be less than or equal to n; P is the transmission packet loss of code stream;

The described threshold value that refreshes is calculated by formula (VII);

$D_{\mathrm{TH}}\left(n\right)=\frac{C_{\mathrm{IR}}}{{\mathrm{\σ}}_{\mathrm{frame}}\left(n\right)}---\left(\mathrm{VII}\right)$

In the formula (VII), C _IRBe fixed value parameter, σ _FrameBe regional movement intensity.Give more to refresh quantity for the big frame of exercise intensity, to obtain better adaptivity.

In order to save the bit resource, the frame refreshing macro block is only selected from even more important moving region.Intra-frame macro block updating mechanism can stop mistake and spread, and has stronger fault-tolerant ability, can further strengthen the fault-tolerant effect of FMO mechanism.

The result of calculation of the macro block updating that flexible macro-block order coding mode that (4) obtains according to step (2) and step (3) obtain forms the encoding scheme of next frame image;

(5) repeating step (2)～(4) are all over until coding.

Experiment effect: adopt the inventive method (AEFMO) that Salesman, Container and three kinds of standard video sequence of Hall are carried out fault-tolerant encoding.Compare with FMO1 method (the FMO Class1 of H.264/AVC stipulating in the standard) and FMO2 method (the FMO type of H.264/AVC stipulating in the standard 2) among the model standard reference model JM15.1 H.264/AVC.The experiment packet loss is 10%, fixed value parameter k, σ ₀, β _H, β _L, σ _{MA_H}, σ _{MA_L}, S _THAnd C _IRBe made as 0.1,100,1.9,1.3,800,300,5 * 10 respectively ⁵With 2 * 10 ⁵Fig. 4 is Y-PSNR (PSNR) experimental result of Salesman sequence, and curve 1,2 and 3 is represented the result of FMO1, FMO2 and three kinds of methods of AEFMO respectively.Fig. 5 is Y-PSNR (PSNR) experimental result of Container sequence, and curve 1,2 and 3 is represented the result of FMO1, FMO2 and three kinds of methods of AEFMO respectively.Fig. 6 is Y-PSNR (PSNR) experimental result of Hall sequence, and curve 1,2 and 3 is represented the result of FMO1, FMO2 and three kinds of methods of AEFMO respectively.Can see that from Fig. 4～6 the present invention is compared to the FMO1 0.03-3.18 decibel (dB) that can improve the quality,, obtain fault-tolerant preferably effect compared to the FMO2 0.23-5.10 decibel (dB) that can improve the quality.

Claims (8)

1. the video fault-tolerant coding method based on self-adaptation flexible macro-block order is characterized in that, may further comprise the steps:

(1) input one two field picture and coding;

(2) moving region of the described image of extraction, the error sensitivity of estimation moving region; According to the error sensitivity of moving region, the flexible macro-block order coding mode that decision-making is optimum;

(3) calculate the accumulation frame difference of all macro block position; When the accumulation frame difference of macro block position in the moving region greater than appointment refresh threshold value the time; Described macro block position is forced frame refreshing when next frame, calculate the quantity and the position of the frame refreshing macro block of moving region thus;

The quantity and the position of the frame refreshing macro block that flexible macro-block order coding mode that (4) obtains according to step (2) and step (3) obtain form the encoding scheme of next frame image;

(5) repeating step (1)～(4) are all over until coding.

2. video fault-tolerant coding method according to claim 1 is characterized in that, in the described step (2), the moving region of the described image of described extraction comprises:

(2.1) the frame difference information of utilizing motion search to produce is carried out initial motion to macro block and is detected, and difference is regarded as the initial motion macro block greater than the macro block of motion detection threshold between incoming frame and the reference frame;

(2.2) analyze the macroblock encoding bit value, described initial motion macro block is carried out filtering, confirm the motion macro block, and the rectangle of confirming to contain the minimum of all motion macro blocks is the moving region.

3. video fault-tolerant coding method according to claim 2 is characterized in that, in the described step (2.1), described motion detection threshold is dynamically adjusted along with the characteristic of not homotactic characteristic and same sequence different frame.

4. video fault-tolerant coding method according to claim 2 is characterized in that, in the described step (2.2), the process that described initial motion macro block is carried out filtering is following:

Calculate the quantity N of initial motion macro block in four peripheral macro blocks of each initial motion macro block _Neb

For N _NebValue is 0 or 1 initial motion macro block, its final motion sign M _FinalBe defined as:

$M_{\mathrm{final}}(x,y,n)=\left\{\begin{array}{cc}1,& \mathrm{if}{B}_{\mathrm{mb}}(x,y,n)\≥[{\mathrm{\λ}}_{\mathrm{filter}}\×B_{\mathrm{avg}}\left(n\right)]\\ 0,& \mathrm{otherwise}\end{array}\right.---\left(\mathrm{III}\right)$

In the formula (III), B _MbRepresent a macroblock encoding bit value, B _AvgRepresent all N _NebThe mean value of initial motion macroblock coding bit value greater than 1; λ _FilterFor the fixed value parameter, for N _NebBe 0 and 1 initial motion macro block, λ _FilterBe made as 1.4 and 0.7 respectively; When the macroblock encoding bit value more than or equal to filtering threshold [λ _Filter* B _Avg(n)] time, final motion sign M _FinalBe 1, be the motion macro block; Otherwise, final motion sign M _FinalBe 0, be non-motion macro block;

For non-initial motion macro block and N _NebValue is greater than other initial motion macro block of 1, M _FinalAnd M _InitialValue equate; M _InitialInitial motion sign for macro block;

Thus, obtain the M of all macro blocks _FinalValue, all M _FinalValue is that 1 macro block is the motion macro block, and the rectangle of containing the minimum of all motion macro blocks is the moving region.

5. video fault-tolerant coding method according to claim 1 is characterized in that, in the described step (2), the error sensitivity of described moving region, adopt and estimate as shown in the formula the calculating formula shown in (IV):

S _MA(n)＝p(n)×T(n)×σ _mb(n)×σ _frame(n) (IV)

In the formula (IV), p is the transmission packet loss of code stream, and T is the texture complexity of moving region, σ _MbBe macro block movement differential in the moving region, σ _FrameBe regional movement intensity.

6. video fault-tolerant coding method according to claim 5 is characterized in that, the texture complexity T of described moving region is the texture complexity t of all macro blocks in the moving region _MbMean value, and the texture complexity t of a macro block _MbCalculating is as shown in the formula shown in (V):

$t_{\mathrm{mb}}(x,y,n)=\frac{\underset{i,j\∈\mathrm{\Ω}(x,y,n)}{\mathrm{\Σ}}{\mathrm{sad}}_{\mathrm{DC}44}(i,j,x,y,n)}{N_{\mathrm{\Ω}}(x,y,n)}---\left(V\right)$

In the formula (V), Ω representes 4 * 4 the set that the neighboring pixel of the left side and top is complete, N _ΩRepresent 4 * 4 quantity in this set, coordinate (i, j) one 4 * 4 positions in its place macro block of expression, sad _DC44It is the difference sum of the neighboring pixel average of 4 * 4 all composition pixels and the left side and top.

7. video fault-tolerant coding method according to claim 1 is characterized in that, in the described step (2), and described error sensitivity according to the moving region, the process of the flexible macro-block order coding mode that decision-making is optimum is following:

The monolithic group coding is all used in non-moving region;

For the moving region, the error sensitivity S when the moving region _MAGreater than the susceptibility threshold S that sets _THThe time, adopt the checker board pattern of two sheet groups to encode; Error sensitivity S when the moving region _MALess than the susceptibility threshold S that sets _THThe time, the smooth pattern-coding of a sheet group of employing.

8. video fault-tolerant coding method according to claim 1 is characterized in that, in the described step (3), the accumulation frame difference of described macro block position is obtained by following formula (VI):

$D_{\mathrm{acc}}(x,y,n)=\underset{t=n_0}{\overset{n}{\mathrm{\Σ}}}p\left(t\right)\×{\mathrm{sad}}_{\mathrm{mv}00}(x,y,t)---\left(\mathrm{VI}\right)$

In the formula (VI), sad _Mv00The coded message that obtains when being 16 * 16 motion search at the zero motion vector place, the difference between reflection incoming frame and the reference frame; n ₀Expression macro block position (x, the image sequence when y) last time carrying out intraframe predictive coding number, n ₀Be less than or equal to n; P is the transmission packet loss of code stream;

The described threshold value that refreshes is:

$D_{\mathrm{TH}}\left(n\right)=\frac{C_{\mathrm{IR}}}{{\mathrm{\σ}}_{\mathrm{frame}}\left(n\right)}---\left(\mathrm{VII}\right)$

In the formula (VII), C _IRBe fixed value parameter, σ _FrameBe regional movement intensity.

Images