CN1688165A - Fast motion assessment method based on object texture - Google Patents

Abstract

A quick motion value-evaluation method of objects includes the following steps: 1, the motion evaluation of border streak macro-blocks: Taking the position determined by MV3 as the center in a reference frame to search for border streak macro-block motion vector within minus or plus sphere, if BAB corresponding to the macro-block is an in-frame encode mode, then MV3=(0,0), 2, the motion evaluation of opaque macro-block in the object includes the selection of the initial search point, determination of search sphere and judgment to the interrupt of the motion evaluation.

Description

The method of object-based texture fast motion assessment

Technical field

What the present invention relates to is a kind of method of estimating motion, and particularly a kind of method of object-based texture fast motion assessment belongs to technical field of video coding.

Background technology

In traditional international video encoding standard, all be to express video sequence with many two-dimensional rectangle frames, whole rectangular frame unification is encoded, do not distinguish concrete object wherein.The expression of this video sequence can not be operated video encoder to certain concrete object in the rectangular frame, has limited the flexibility of video coding greatly.In object-based video encoding standard, the coding of rectangular frame is changed into coding to the object video of arbitrary shape, independent behavior, independent texture.Because the introducing of object video notion need be encoded to three category informations of object video when carrying out video coding: shape coding, motion encoded, texture coding.Texture information in the video object plane (VOP) has been divided into many 16 * 16 macro block, and macro block is divided three classes: transparent macro block, opaque macro block, boundary macroblocks.In order to reduce the temporal redundancy between the video sequence, improve code efficiency, when encoding between conducting frame, must carry out estimating motion and compensation (ME/MC) to the texture information of object video: in reference texture information, find the best matching blocks (corresponding matching error minimum) of present encoding texture macro block, the residual block after its motion compensation is encoded.In the estimating motion method, global search (FS) method is by the whole candidate's reference blocks in the match search zone, find best match block, therefore global search also is accurate movement estimation method, but this method need take the amount of calculation of whole texture encoder more than 80%, becomes the highest module of computation complexity in the encoder.

Find by prior art documents, people such as Shan Zhu are at IEEE Transactions on ImageProcessing, vol.9, no.2, pp.287-290, Feb.2000 (IEEE image processing transactions, the 9th volume, the 2nd phase, the 287-290 page or leaf, in February, 2000) article of delivering on " A New Diamond Search Algorithmfor Fast Block-Matching Motion Estimation " (" a kind of new diamond search algorithm that is used for quick block matching motion valuation ") has proposed a kind of new fast motion assessment algorithm, is called diamond search (DS) algorithm.This algorithm is by two search window that vary in size, and seeks the point with minimum movement compensation, this point corresponding final motion vector.This algorithm has two obvious defects, a defective is the style characteristic that this algorithm does not utilize object video, be under coding environment, to propose based on frame, if just simply this fast motion assessment method is applied in the object-based video coding, sometimes can influence the effect of estimating motion because of the border characteristic of object video; Another defective is that this algorithm supposes that overall smallest point is around null vector, and in fact not all video sequence can both satisfy this hypothesis, for the sequence that can not satisfy hypothesis, diamond search (ds) is easy to be absorbed in local smallest point, thereby makes the deleterious of estimating motion.

Summary of the invention

The present invention is directed to the prior art above shortcomings, propose a kind of method (OBS) of object-based texture fast motion assessment.Make the matching properties of its shape information that has not only made full use of object video, target edges macro block; And made full use of the correlation that the object intra-macroblock moves, be absorbed in local smallest point when having avoided estimating motion; Made full use of texture macroblock encoding characteristic simultaneously.The present invention is better than existing various fast motion assessment method, and matching error keeps close, and this method and existing method be complementary, and the best searching route that can utilize existing method to propose is with the speed of further raising estimating motion.

The present invention is achieved by the following technical solutions, and concrete steps are as follows:

(1) estimating motion of border texture macro block

Because in object-based coding, boundary macroblocks has appearred in the edge of object, and opaque macro block matching properties when carrying out estimating motion of boundary macroblocks and object inside is different, so the present invention treats with a certain discrimination border texture macro block and opaque texture macro block.For the border texture macro block, core concept of the present invention can be described below: owing in object-based coding, before the encoding texture macro block, carry out the coding of corresponding shape BAB piece (two-value shape piece) earlier.When coding BAB piece, calculated the shape motion vector MVs of BAB piece.Because boundary B AB piece and border texture macro block are one to one, its movable information is closely similar, therefore when carrying out the estimating motion of border texture macro block, the MVs of BAB piece when making full use of shape coding, with the predicted value of MVs, thereby improve the speed of estimating motion greatly as border texture macro block motion vector.

In the present invention, the estimating motion method of border texture macro block is: when carrying out shape coding, the shape motion vector MVs that keeps corresponding BAB piece, the position of determining with MVs in reference frame is the center, the motion vector MV of search border texture macro block in ± 1 scope.If with the corresponding BAB piece of border texture macro block be (Intra) coding mode in the frame, then make MVs=(0,0).

The search point of the present invention during with estimating motion weighed the speed of estimating motion, with the absolute value of border texture macro block motion compensation difference and the polygon matching error of average MAE when weighing border texture macro block estimating motion, be shown below:

$\mathrm{MAE}(m,n)=\frac{1}{M}\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{N-1}{\mathrm{\Σ}}}|f_k(i,j)-f_{k-1}(i+m,j+n)|\×({\mathrm{Alpha}}_{\mathrm{original}}(i,j)\≠0)$

N represents the size of macro block in the formula, and M represents the number of opaque pixel in the macro block, and (m n) is motion vector, f _k(i, the j) data of expression current macro, f _K-1(i, the j) data of expression motion compensation macro block, Alpha _Original(i j) is macro block corresponding shape data.

Compare with the global search method, the search point when the inventive method greatly reduces border texture macro block estimating motion has only 0.83% of former search point, and the matching error maintenance is close.

(2) estimating motion of the opaque texture macro block of object inside

If current macro is the opaque texture macro block of object inside, then to carry out conventional estimating motion.Though this is because opaque BAB piece and opaque texture macro block also are one to one, but their motion vector has very big difference, because the BAB piece is mainly considered contours of objects information when carrying out estimating motion, opaque BAB piece is contiguous any one opaque BAB piece in the match reference Alpha plane probably.Therefore opaque texture macro block still will carry out conventional estimating motion.

In the present invention, this estimating motion process is described below from three aspects of judgement of determining and interrupting the estimating motion process of the choosing of initial search point, hunting zone respectively:

1. initial search point chooses

When in the present invention, the opaque texture macro block of object inside carries out estimating motion initial search point choose and searching method is: the intermediate value MV that calculates the motion vector of left side macro block, the top macroblock of current macro, upper right macro block _Median

2. the hunting zone determines

In order to accelerate the speed of estimating motion, in the present invention, the searching method of the opaque texture macro block of object inside is: in reference frame, with MV _MedianThe position of determining is the center, search MV in ± 4 scopes.So just can significantly reduce search point, because the MVD overwhelming majority is all in ± 4 scope, so this method can't obviously increase matching error.The simulation experiment result shows that search point of the present invention has only 8.33% of conventional method search point, and matching error does not almost increase.

For the selection of best searching route, the best searching route that can utilize existing fast motion assessment method to propose.The front has been set forth and has been adopted the present invention can not be absorbed in the local smallest point of matching error, so utilize other fast motion assessment method can not influence the effect of estimating motion.Therefore, the present invention and existing fast motion assessment method are complementary, and optimum operating mode is the associating use.For example: the present invention can adopt fast methods such as diamond search, the search of four steps, new three step search in ± 4 hunting zones.Can unite other estimating motion method of use is advantage of the present invention.

3. interrupt the judgement of estimating motion process

In the estimating motion process, if the coupling between current macro and candidate's the reference macroblock satisfies after the certain precision, can interrupt the estimating motion process, need not to continue search, search point can be reduced like this, and matching error can't be obviously increased.

In the present invention, use two kinds of determination methods of interrupting the estimating motion process:

First kind: if the left side macro block of current macro, top macroblock, upper right macro block all are interframe encodes, and the motion vector of three macro blocks is identical, then interrupts the estimating motion process, and the motion vector of current macro equals the motion vector of adjacent macroblocks.The simulation experiment result shows that search point has reduced much when adopting this interruption determination methods, thereby has accelerated the process of estimating motion, and matching error is not significantly increased.

Second kind: the matching error SAD of four pieces in the difference computing macro block _{8 * 8}Value is if four pieces all satisfy SAD _{8 * 8}＜20 * QP, then can judge the residual error macro block that carries out after the motion compensation quantize through DCT-after for complete 0, interrupt the estimating motion process.Also adopt this interruption determination methods when boundary macroblocks is carried out estimating motion, wherein SAD be the macro block motion compensated residual absolute value and.

When the present invention weighed the matching error of macro block in the estimating motion process with SAD, SAD was provided by following formula:

$\mathrm{SAD}(m,n)=\underset{i=0}{\overset{N}{\mathrm{\Σ}}}\underset{j=0}{\overset{N}{\mathrm{\Σ}}}|f_k(i,j)-f_{k-1}(i+m,j+n)|\×({\mathrm{Alpha}}_{\mathrm{original}}(i,j)\≠0)$

N represents the size of macro block, f in the formula _k(i, the j) data of expression current macro, f _K-1((m n) is motion vector, Alpha for i, the j) data of expression motion compensation macro block _Original(i, j) expression current macro corresponding shape information.

The simulation experiment result shows, along with the increase of quantization parameter, search point is in continuous minimizing, thereby accelerated the speed of estimating motion, and this is very useful in the application of low code check; Though this moment, matching error increased some, did not influence whole code efficiency, because the motion compensated residual of this moment finally is quantified as 0, did not encode.

No matter be the Rule of judgment that satisfies above-mentioned a certain interruption estimating motion process, still searched last search point, the estimating motion process all can stop, and obtains final motion vector MV.

Compare with existing texture fast motion assessment algorithm, the inventive method has made full use of the shape information of object video and the matching properties of target edges macro block, and this also is a core concept of the present invention.The present invention adopts the initial value of the shape motion vector of boundary shape piece as border texture macro block motion vector, and then search further obtains final motion vector, thereby has greatly improved the estimating motion speed of border texture macro block.The present invention has also made full use of the correlation of object intra-macroblock motion, with around close on the macro block motion vector the search center of intermediate value during as opaque texture macro block estimating motion, thereby make the final motion vector of initial search point distance nearer, motion compensation error is absorbed in local smallest point when not only having avoided estimating motion, and can in littler hunting zone, search for final motion vector, motion compensation error does not have tangible increase.In addition, the present invention has also made full use of texture macroblock encoding characteristic, further improves the speed of estimating motion by the determination methods of two interruption estimating motion processes.

Description of drawings

Fig. 1 is a flow chart of the present invention

Fig. 2 is the adjacent macroblocks distribution schematic diagram of current macro in the inventive method

Embodiment

Content below in conjunction with accompanying drawing and method provides embodiments of the invention.

As shown in Figure 1, the present invention can be divided into following step.

(1) estimating motion of border texture macro block

In the present invention, judge at first whether current macro is the border texture macro block, if border texture macro block, then be: when carrying out shape coding to its estimating motion method of carrying out, the shape motion vector MVs that keeps corresponding BAB piece, the position of determining with MVs in reference frame is the center, the motion vector MV of search border texture macro block in ± 1 scope.If with the corresponding BAB piece of border texture macro block be (Intra) coding mode in the frame, then make MVs=(0,0).

The operation result of border texture macro block estimating motion relatively (hunting zone is ± 15) when table 1 adopts different searching method

The sequence name	Global search (FS) method			Object-based search (OBS) method
							??MAE	Search point	??％	??MAE	Search point	????％
	????News	??4.25	??5856495	??100	??4.62	??48805							??0.83
????Container							??7.68	??5630182	??100	??7.89	??46928	??0.83
	????Coastguard	??10.76	??5035014	??100	??11.42	??41970							??0.83
????Stefan							??12.55	??2618482	??100	??13.69	??22473	??0.86

Table 1 has provided when adopting global search (FS) method and the inventive method (OBS) respectively, and the operation result of border texture macro block estimating motion relatively.Cycle tests is 100 frames, CIF (352 * 288) form.The search point of the present invention during with estimating motion weighed the speed of estimating motion, with the absolute value of border texture macro block motion compensation difference and average (MAE) polygon matching error when weighing border texture macro block estimating motion, be shown below:

$\mathrm{MAE}(m,n)=\frac{1}{M}\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{N-1}{\mathrm{\Σ}}}|f_k(i,j)-f_{k-1}(i+m,j+n)|\×({\mathrm{Alpha}}_{\mathrm{original}}(i,j)\≠0)$

N represents the size of macro block in the formula, and M represents the number of opaque pixel in the macro block, and (m n) is motion vector, f _k(i, the j) data of expression current macro, f _K-1(i, the j) data of expression motion compensation macro block, Alpha _Original(i j) is macro block corresponding shape data.

As can be seen from Table 1, compare with the global search method, the search point when adopting the OBS method to greatly reduce border texture macro block estimating motion has only 0.83% of former search point, and the matching error maintenance is close.

(2) estimating motion of the opaque texture macro block of object inside

In the present invention, if current macro is not the border texture macro block, be that current macro is the opaque texture macro block of object inside, then the estimating motion process that this macro block is carried out comprises three steps of judgement of determining and interrupting the estimating motion process of the choosing of initial search point, hunting zone.

1. initial search point chooses

Initial search point choose and searching method is: the intermediate value MV that calculates the motion vector of left side macro block, the top macroblock of current macro, upper right macro block _Median, the distribution of adjacent macroblocks as shown in Figure 2, in reference frame with MV _MedianThe position of determining is the center, beginning searching moving vector M V.

Table 2 adjacent macroblocks MV gets the method for intermediate value

The adjacent macroblocks number	The type of coding of adjacent macroblocks	???MV median
			????1	????Inter	The MV of adjacent macroblocks
????Intra	???(0，0)
		????2		????Intra，Intra	???(0，0)
????Inter，Inter	Intermediate value
			????Inter，Intra	The MV of Inter coded macroblocks
????3	????Intra，Intra，Intra				???(0，0)
		????Inter，Inter，Inter	Intermediate value
	????Inter，Inter，Intra			Intermediate value

Inter，Intra，Intra

The MV of Inter coded macroblocks

Intermediate value MV _MedianComputational methods as shown in table 2, can be described below:

If current macro has only an adjacent macroblocks, then

If this adjacent macroblocks is interframe (Inter) coding, then MV _MedianGet the MV that makes this adjacent macroblocks,

If this adjacent macroblocks is (Intra) coding, then MV in the frame _MedianGet work (0,0);

If current macro has two adjacent macroblocks, then

If two adjacent macroblocks all are intraframe coding, then MV _MedianGet work (0,0),

If two adjacent macroblocks all are interframe encode, then MV _MedianGet the intermediate value of making adjacent macroblocks MV,

If two one of adjacent macroblocks are intraframe codings, another is an interframe encode, then MV _MedianGet the MV that makes this inter-coded macroblocks;

If current macro has three adjacent macroblocks, then

If three adjacent macroblocks all are intraframe coding, then MV _MedianGet work (0,0),

If three adjacent macroblocks all are interframe encode, then MV _MedianGet the intermediate value of making adjacent macroblocks MV,

If one is intraframe coding in three adjacent macroblocks, two is interframe encode in addition, then MV _MedianGet the intermediate value of making adjacent macroblocks MV,

If one is interframe encode in three adjacent macroblocks, two is intraframe coding in addition, then MV _MedianGet the MV that makes this inter-coded macroblocks;

Need to prove: when getting intermediate value,, then make the MV=(0,0) of the 3rd adjacent macroblocks if having only two adjacent macroblocks; If three adjacent macroblocks are arranged but wherein have one to be intraframe coding, then make the MV=(0,0) of inter-coded macroblocks.When table 3 had provided respectively intermediate value with null vector (0,0) and adjacent macroblocks motion vector as search center, the operation result of opaque texture macro block estimating motion relatively.Cycle tests is 100 frames, CIF (352 * 288) form.As can be seen from Table 3, when adopting method of the present invention to carry out the estimating motion of opaque texture macro block, matching error does not have obvious variation, illustrates that method of the present invention is feasible.

The operation result of opaque texture macro block estimating motion relatively (hunting zone is ± 15) when table 3 is chosen different initial search point

The sequence name	With (0,0) is the center	With MV medianBe the center
			????MAE	????MAE
	??News	????3.51			????3.51
??Container			????7.43	????7.43
	??Coastguard	????7.51			????7.52
??Stefan			????11.06	????11.04

2. the hunting zone determines

In the present invention, in order to accelerate the speed of estimating motion, the searching method of opaque texture macro block is: in reference frame, with MV _MedianThe position of determining is the center, search MV in ± 4 scopes.So just can significantly reduce search point, because the MVD overwhelming majority is all in ± 4 scope, so this method can't obviously increase matching error.The operation result that table 4 has provided opaque texture macro block estimating motion when adopting conventional method and the inventive method compares.Cycle tests is 100 frames, CIF (352 * 288) form.As can be seen from the table, search point of the present invention has only 8.33% of conventional method search point, and matching error does not almost increase.

The operation result of opaque texture macro block estimating motion relatively when table 4 adopted different searching method

The sequence name	With (0,0) is the center, hunting zone ± 15			With MVmedian is the center, hunting zone ± 4
							????MAE	Search point	??％	????MAE	Search point	????％
	??News	????3.51	??10566720	??100	????3.52	????880560							????8.33
??Container							????7.43	??2768640	??100	????7.43	????230720	????8.33
	??Coastguard	????7.51	??1084800	??100	????7.52	????90400							????8.33
??Stefan							????11.06	??529765	??100	????11.09	????44160	????8.34

3. interrupt the judgement of estimating motion process

In the present invention, first kind of determination methods of interrupting the estimating motion process is: if the left side macro block of current macro, top macroblock, upper right macro block all are interframe encodes, and the motion vector of three macro blocks is identical, then interrupt the estimating motion process, the motion vector of current macro equals the motion vector of adjacent macroblocks.The operation result that table 5 has provided opaque texture macro block estimating motion when adopting global search and first kind of interruption determination methods of the present invention respectively compares.Cycle tests is 100 frames, CIF (352 * 288) form.As can be seen from Table 5, search point has reduced much when adopting the present invention to interrupt determination methods, thereby has accelerated the process of estimating motion, and matching error is not significantly increased.

The operation result of opaque texture macro block estimating motion relatively when table 5 adopted global search and first kind of interruption determination methods

The sequence name	Global search (FS)			When equating, adjacent macroblocks MV interrupts search
							MAE	Search point	??％	??MAE	Search point	????％

??News	??3.51	??10566720	??100	??3.54	??766080	??7.25
							??Container	??7.43	??2768640	??100	??7.64	??358080	??12.93
??Coastguard	??7.51	??1084800	??100	??7.54	??227520	??20.97
							??Stefan	??11.06	??529765	??100	??11.12	??496320	??93.69

In the present invention, second kind of determination methods of interrupting the estimating motion process is: the matching error SAD of four pieces in the difference computing macro block _{8 * 8}Value is if four pieces all satisfy SAD _{8 * 8}＜20 * QP, then can judge the residual error macro block that carries out after the motion compensation quantize through DCT-after for complete 0, interrupt the estimating motion process.When being carried out estimating motion, also adopts the border texture macro block this interruption determination methods.Table 6 has provided when adopting global search and second kind of interruption of the present invention determination methods respectively, and the operation result of border texture macro block and opaque texture macro block estimating motion relatively.As can be seen from the table, along with the increase of quantization parameter, search point is in continuous minimizing, thereby accelerated the speed of estimating motion, and this is very useful in the application of low code check; Though this moment, matching error increased some, did not influence whole code efficiency, because the motion compensated residual of this moment finally is quantified as 0, did not encode.

The operation result of whole macro block estimating motions compared (100 frame CIF sequence) when table 6 adopted global search and second kind of interruption determination methods

The sequence name	Global search (FS)			Interrupt search during SAD8 * 8＜20*QP
													???????????QP＝10			???????????QP＝15			???????????QP＝18
				?MAE	Search point	??％	MAE	Search point	?％	?MAE	Search point	?％										?MAE	Search point	?％
	??News	?3.78	?16423215										?100	?3.90	?12678464	?77.20	?4.57	?10930243	?66.55	?4.80	?10386289				?63.24
??Container				?7.60	?8398822	?100	?7.79	?7617152	?90.69	?9.20	?7322507	?87.18										?9.66	?7179155	?85.48
	??Coastguard	?10.18	?6119814										?100	?10.44	?5694496	?93.05	?12.14	?5601431	?91.53	?13.09	?5545856				?90.62
??Stefan				?12.30	?3148247	?100	?13.01	?2910883	?92.46	?14.41	?2809115	?89.23										?14.94	?2760534	?87.68

No matter be the Rule of judgment that satisfies above-mentioned a certain interruption estimating motion process, still searched last search point, the estimating motion process all can stop, and obtains final motion vector MV.

Estimating motion speed of the present invention is greatly faster than the estimating motion speed of global search method, and is better than other fast motion assessment method, and that matching error keeps is close.In order to contrast, provided the experimental result that global search method (FS), diamond search method (DS), the present invention propose here based on object search method (OBS) and OBS+ diamond search method of the present invention.Use the MPEG-4 standard test sequences of QCIF (176 * 144) and CIF (352 * 288) form, the length of each sequence is 100 frames, first frame is the IVOP coding mode, and back 99 frames all are that the PVOP coding mode (is IPPP ...), the hunting zone of macro block is ± 16.

The essential characteristic of cycle tests is: " Akiyo " and " Weather " is the QCIF form, and " News ", " Container ", " Coastguard " and " Stefan " are the CIF form; Wherein " Container ", " Akiyo " and " News " are the microinching sequence, and " Coastguard " and " Weather " is the middle motion sequence, and " Stefan " is the rapid movement sequence.

With the absolute value of macro block motion compensation difference and average (MAE) matching error when weighing the macro block estimating motion, the search point during with estimating motion is weighed the speed of estimating motion.Table 7 has provided the statistics of emulation experiment.

As can be seen from Table 7, the MAE of the MAE of OBS method and FS, DS method correspondence is very approaching, but it is big a little, mainly be because: interrupt in the determination methods at second kind of the OBS method, when the residual block of the match block correspondence of judging the candidate quantizes the back for complete zero piece through DCT-, with the stop motion valuation, corresponding MAE will be bigger this moment, but do not influence final code efficiency, because complete zero piece is not encoded; In addition, the border texture macro block utilizes the motion vector of its corresponding shape piece to search for the texture motion vector then as predicted value in smaller scope, also can increase MAE slightly.

It can also be seen that from table 7 compare with other fast motion assessment method, search speed of the present invention is the fastest.Because the present invention and other fast motion assessment method are complementary technically, so when searching moving is vectorial in the OBS method is united the scope of other fast motion assessment method (for example diamond search method) ± 4, search point can further reduce, and promptly the speed of estimating motion can further be accelerated.

The operation result of the various estimating motion methods of table 7 is (cycle tests 100 frames) relatively

	The sequence name	???Akiyo	???Weather	????News	?Container	?Coastguard	????Stefan
								Form	???QCIF	???QCIF	????CIF	????CIF	????CIF	????CIF
	Global search (FS)	????MAE	???4.84	???7.22	????3.78	????7.60	????10.18								????12.30
Search point								???4437801	???3227518	????16423215	????8398822	????6119814	????3148247
		????％	???100	???100	????100	????100	????100							????100
Diamond search (DS)								????MAE	???4.95	???7.35	????3.88	????7.71	????10.30		????13.35
	Search point	???55766	???41292	????206305	????105608	????83130	????67773
								????％	???1.26	???1.28	????1.26	????1.26	????1.36	????2.15
	Of the present invention based on object search (OBS)	????MAE	???5.10	???7.59	????3.96	????7.85	????10.69								????13.77
Search point								???25759	???23122	????107311	????61629	????49096	????46708
		????％	???0.58	???0.72	????0.65	????0.73	????0.80							????1.48
								????MAE	???5.10	???7.58	????3.96	????7.86	????10.70		????13.73

OBS+ diamond search of the present invention	????MAE	????5.10	????7.58	????3.96	????7.86	????10.70	????13.73
								Search point	????18728	????18544	????42593	????46113	????41423	????28823
	????％	????0.42	????0.57	????0.26	????0.55	????0.68	????0.92

Can draw from table 7 as drawing a conclusion: the effect that (1) the inventive method is carried out estimating motion to various sequences is all fine; (2) the estimating motion speed of the inventive method is better than the estimating motion speed of global search method greatly, and is better than other fast motion assessment method, and the matching error maintenance is close; (3) when the inventive method is united other fast motion assessment method when (for example diamond search method), its average search point can further reduce, the speed that is estimating motion can further be accelerated, therefore the inventive method (OBS) is united the operational mode that other fast motion assessment method is the best, can make full use of the advantage of each method like this.

Compare with existing texture fast motion assessment algorithm, advantage of the present invention is: (1) has made full use of the shape information of object video and the matching properties of target edges macro block; (2) adopt the initial value of the shape motion vector of boundary shape piece, and then search further obtains final motion vector, thereby greatly improved the estimating motion speed of border texture macro block as border texture macro block motion vector; (3) made full use of the correlation that the object intra-macroblock moves, with around close on the macro block motion vector the search center of intermediate value during as opaque texture macro block estimating motion, thereby make the final motion vector of initial search point distance nearer, motion compensation error is absorbed in local smallest point when not only having avoided estimating motion, and can in littler hunting zone, search for final motion vector, motion compensation error does not have tangible increase; (4) also make full use of texture macroblock encoding characteristic, further improved the speed of estimating motion by the determination methods of two interruption estimating motion processes.

Claims (4)

1, a kind of method of object-based texture fast motion assessment is characterized in that, step is as follows:

(1) estimating motion of border texture macro block

When carrying out shape coding, keep the shape motion vector MVs of corresponding BAB piece, the position of determining with MVs in reference frame is the center, the motion vector MV of search border texture macro block in ± 1 scope, if with the corresponding BAB piece of border texture macro block be intra-frame encoding mode, then make MVs=(0,0);

(2) estimating motion of the opaque texture macro block of object inside

1. choosing of initial search point: the intermediate value MV that calculates the motion vector of left side macro block, the top macroblock of current macro, upper right macro block _Median

2. the hunting zone is definite: in reference frame, with MV _MedianThe position of determining is the center, search MV in ± 4 scopes;

3. interrupt the judgement of estimating motion process: comprise two kinds of methods, specific as follows:

First kind: if the left side macro block of current macro, top macroblock, upper right macro block all are interframe encodes, and the motion vector of three macro blocks is identical, then interrupts the estimating motion process, and the motion vector of current macro equals the motion vector of adjacent macroblocks;

Second kind: the matching error SAD of four pieces in the difference computing macro block _{8 * 8}Value is if four pieces all satisfy SAD _{8 * 8}＜20 * QP, then judge after the residual error macro block carry out after the motion compensation quantizes through DCT-and be complete 0, interrupt the estimating motion process, also adopt this interruptions determination methods when boundary macroblocks is carried out estimating motion, wherein SAD be the macro block motion compensated residual absolute value with;

No matter be the Rule of judgment that satisfies above-mentioned a certain interruption estimating motion process, still searched last search point, the estimating motion process all can stop, and obtains final motion vector MV.

2, the method for object-based texture fast motion assessment according to claim 1 is characterized in that, the initial search point of described step (2) choose the intermediate value MV of the motion vector of the left side macro block of current macro, top macroblock, upper right macro block _MedianComputational methods as follows:

If current macro has only an adjacent macroblocks, then

As if this adjacent macroblocks is interframe encode, then MV _MedianGet the MV that makes this adjacent macroblocks,

As if this adjacent macroblocks is intraframe coding, then MV _MedianGet work (0,0);

If current macro has two adjacent macroblocks, then

If two adjacent macroblocks all are intraframe coding, then MV _MedianGet work (0,0),

If two adjacent macroblocks all are interframe encode, then MV _MedianGet the intermediate value of making adjacent macroblocks MV,

If two one of adjacent macroblocks are intraframe codings, another is an interframe encode, then MV _MedianGet the MV that makes this inter-coded macroblocks;

If current macro has three adjacent macroblocks, then

If three adjacent macroblocks all are intraframe coding, then MV _MedianGet work (0,0),

If three adjacent macroblocks all are interframe encode, then MV _MedianGet the intermediate value of making adjacent macroblocks MV,

If one is intraframe coding in three adjacent macroblocks, two is interframe encode in addition, then MV _MedianGet the intermediate value of making adjacent macroblocks MV,

If one is interframe encode in three adjacent macroblocks, two is intraframe coding in addition, then MV _MedianGet the MV that makes this inter-coded macroblocks;

Need to prove: when getting intermediate value,, then make the MV=(0,0) of the 3rd adjacent macroblocks if having only two adjacent macroblocks; If three adjacent macroblocks are arranged but wherein have one to be intraframe coding, then make the MV=(0,0) of inter-coded macroblocks.

3, the method for object-based texture fast motion assessment according to claim 1, it is characterized in that, described second kind determination methods of interrupting the estimating motion process, weigh the matching error of macro block in the estimating motion process with SAD, SAD is provided by following formula:

$\mathrm{SAD}(m,n)=\underset{i=0}{\overset{N}{\mathrm{\Σ}}}\underset{j=0}{\overset{N}{\mathrm{\Σ}}}|f_k(i,j)-f_{k-1}(i+m,j+n)|\×(\mathrm{Alph}{a}_{\mathrm{original}}(i,j)\≠0)$

N represents the size of macro block, f in the formula _k(i, the j) data of expression current macro, f _K-1((m n) is motion vector, Alpha for i, the j) data of expression motion compensation macro block _Original(i, j) expression current macro corresponding shape information.

4, the method for object-based texture fast motion assessment according to claim 1, it is characterized in that, in the described step (1), search point during with estimating motion is weighed the speed of estimating motion, with the absolute value of border texture macro block motion compensation difference and the polygon matching error of average MAE when weighing border texture macro block estimating motion, be shown below:

$\mathrm{MAE}(m,n)=\frac{1}{M}\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{N-1}{\mathrm{\Σ}}}|f_k(i,j)-f_{k-1}(i+m,j+n)|\×(\mathrm{Alph}{a}_{\mathrm{original}}(i,j)\≠0)$

N represents the size of macro block in the formula, and M represents the number of opaque pixel in the macro block, and (m n) is motion vector, f _k(i, the j) data of expression current macro, f _K-1(i, the j) data of expression motion compensation macro block, Alpha _Original(i j) is macro block corresponding shape data.

Images