CN1622635A - Motion estimating method and device in video compression system - Google Patents

Abstract

The motion estimating and measuring method and device in video compression system may adopts cost function unlike that in available technology in calculating one optimal motion vector of the current coded block. The cost function the present invention adopts has not only the traditional 'absolute difference sum' considered, but also the quantizing step length and the pixel difference value variation state of the current coded block relative to the blocks in one search arean of the destination picture considered, so that the cost function the present invention can find out the optimal motion vector corresponding to residual value capable of being treated to obtain relative high compression efficiency.

Description

In video compression system, carry out the method and the relevant apparatus of moltion estimation

Technical field

The present invention relates to a kind of method and relevant apparatus that a video compression system carries out moltion estimation that be used for, the cost function that particularly a kind of use is different from known technology carries out the method and the relevant apparatus of moltion estimation.

Technical background

In recent years, the multimedia technology vigorous growth, and, have more and more multiple video compression standard to be suggested successively along with Development of Multimedia Technology.For instance, the MPEG technology of various version promptly is to be used for storing and the compression standard of transmitting video data with digital form.Then also have ITU H.261, H.263 and ISO 10918 or the like as for other video compression standard.

The international video standard that MPEG proposed has defined a kind of specification that is used for compressed digital video.Dynamic video is made up of a series of picture (picture), and each picture can be regarded the two-dimensional array of being made up of many pixels (pixel) as, and each picture can also be called a single frame (frame) in the dynamic video.And in the international video standard that MPEG proposed, be to have defined four kinds of different types of pictures, be respectively: the I picture, not encoding with reference to other picture draws; The P picture, with reference to before I picture or P picture, encoding via moltion estimation (motion estimation) draws; The B picture, with reference to before and I picture afterwards or P picture, encoding via moltion estimation draws; And the D picture, search among the pattern (fast forward search mode) before being used in fast.

And the video compression system that uses above-mentioned various international video standards all can use usually based on the block (block) or the moltion estimation technology of macro zone block (macroblock), to eliminate temporal repeatability (temporal redundancy).When carrying out moltion estimation, for the current encoder block in the present picture (current picture), video compression system can be found out a most identical block (bestmatching block) the most similar to this current encoder block from a target picture (target picture).This moment is for the current encoder block, video compression system only need store motion vector (motion vector) and the residual value (residual) that (or transmission) calculates, and can represent the information (residual value is the difference situation that is used for representing pixel value between current encoder block and the most identical block) that is comprised in this current encoder block.

And in known technology; when the video compression system desire is looked for one during identical block in a Search Area (search range); usually can use the cost function (cost function) of a kind of being called " antipode summation " (sum ofabsolute difference), as shown below:

$\mathrm{SAD}(x,y)=\underset{i=i0}{\overset{i1}{\mathrm{\Σ}}}\underset{j=j0}{\overset{j1}{\mathrm{\Σ}}}|C_{i,j}-P_{i+x,j+y}|$

Wherein, (x is an interior candidate motion vector of Search Area y), (i1-i0) * (j1-j0) is the size of current encoder block, C _{I, j}Be the pixel in the current encoder block, P _{I+x, j+y}It then is the pixel in the Search Area in the target picture.

The video compression system of known technology can be from Search Area, find out one can so that above-mentioned cost function reach minimum value a candidate motion vector (x, y), as the suitableeest motion vector of current encoder block (x1, y1).Such practice mainly is in order to find out the most identical block with Minimum Residual residual value, can to make thus to reach compression degree preferably for residual value.But the suitableeest motion vector (x1 that finds out with said method, y1) but may not be certain to reach compression degree preferably, so U.S. the 5th, 847, No. 776 patent case has proposed another kind of cost function, in seeking the process of the suitableeest motion vector, except considering above-mentioned antipode summation, also list the size of motion vector in consideration, can between the suitableeest motion vector of finding out and corresponding residual value, average out thus.

Yet, a variety of video compression systems are arranged when residual value is compressed, discrete cosine transform is used in the capital, and (residual value that will be positioned at space field (spatialdomain) is converted to frequency field (frequency domain) for discrete cosine transform, algorithm DCT).Use corresponding quantization matrix (quantization matrix) then and can carry out the quantification program to the residual value that is positioned at frequency field with quantization step (quantization step) Qp of selected bit rate (bit rate) change of system.Because the matrix after quantizing is a two-dimensional matrix, so the 2-D data that system also can use the saw chi to scan (zig-zagscan) or the staggered mode that scans (alternate scan) after will quantizing is scanned into the data of one dimension, carries out the computing of variable length code (Variable length coding) at last again.

When carrying out the computing of variable length code, if it is more little to be positioned at the frequency distribution scope of residual value of frequency field, the code length (code length) of the coding residual value of coming out usually will short more (this have also just represented for residual value and has reached compression degree preferably) at last.But be to use the method that patent case proposed of No. the 5th, 847,776, the known technology or the U.S., the way that has that the most identical block of being found out loses makes the residual value that is positioned at frequency field have little frequency distribution scope.Even the most identical block of finding out this moment can be so that the residual value on the space field has minimum antipode summation, but scan after the computing of (or other the mode that scans) and variable length code or the like through discrete cosine transform, quantification program, saw chi, resulting residual value but may not be certain to have the shortest code length, so may not be certain the compression effectiveness that can reach best.This is the subject matter that known technology faces.

Summary of the invention

Therefore the method and the relevant apparatus of the main purpose of the present invention cost function (cost function) that is to provide a kind of use to be different from known technology are to solve the problem that above-mentioned known technology is faced.

According to a claim of the present invention, be to disclose a kind of method that a video compression system carries out moltion estimation that is used for, this video compression system is divided into a plurality of blocks with a present picture, this method includes following steps: for the current encoder block in this present picture, progressively inspect a plurality of candidate motion vector (x in the Search Area, y), wherein this current encoder block includes (i1-i0) * (j1-j0) individual pixel; (x y), calculates the value of a cost function for each candidate motion vector; And with can make in this Search Area this cost function draw minimum value candidate motion vector (x, y) be made as this current encoder block the suitableeest motion vector (x1, y1).This cost function as shown below

$\mathrm{CF}(x,y)=\underset{i=i0}{\overset{i1}{\mathrm{\Σ}}}\underset{j=j0}{\overset{j1}{\mathrm{\Σ}}}\left\{\right|C_{i,j}-P_{i+x,j+y}|+h(\mathrm{Qp},C_{i,j},P_{i+x,j+y})\}$

C wherein _{I, j}Be the pixel in this current encoder block, P _{I+x, j+y}Be the pixel in this Search Area in the target picture, Qp then is a quantization step.

According to another claim of the present invention, be to disclose a kind of device that a video compression system carries out moltion estimation that is used for, this video compression system is divided into a plurality of blocks with a present picture, for the current encoder block in this present picture, this device can be in a Search Area a plurality of candidate motion vector (x, y) determine a suitableeest motion vector (x1 in, y1), wherein include (i1-i0) * (j1-j0) individual pixel in this current encoder block, this device includes: an antipode computing unit is used for calculating | C _{I, j}-P _{I+x, j+y}| value, C wherein _{I, j}Be the pixel in this current encoder block, P _{I+x, j+y}It is the pixel in this Search Area in the target picture; One spatial variations computing module is used for calculating function h (Qp, a C _{I, j}, P _{I+x, j+y}) value, wherein Qp is a quantization step; An and motion vector decision module, be coupled in this antipode computing unit and this spatial variations computing module, be used for calculating the value of a cost function, and this cost function that makes that this Search Area is interior draws the candidate motion vector (x of minimum value, y) be set at this current encoder block the suitableeest motion vector (x1, y1).This cost function as shown below

$\mathrm{CF}(x,y)=\underset{i=i0}{\overset{i1}{\mathrm{\Σ}}}\underset{j=j0}{\overset{j1}{\mathrm{\Σ}}}\left\{\right|C_{i,j}-P_{i+x,j+y}|+h(\mathrm{Qp},C_{i,j},P_{i+x,j+y})\}$

An advantage of the present invention is, because employed cost function can be subjected to this current encoder block with respect to a candidate motion vector (x, y) changing condition of the value differences of the block that is corresponded to influences, so use method proposed by the invention and relevant apparatus can guarantee that the corresponding residual value of the suitableeest motion vector that calculates can have compression efficiency preferably after handling.

The accompanying drawing summary

Fig. 1 is the embodiment flow chart of method proposed by the invention.

Fig. 2 is the embodiment functional block diagram of device proposed by the invention.

Graphic symbol description

200 devices

220 antipode computing units

240 spatial variations computing modules

242 first computing units

244 second computing units

246 multipliers

260 motion vector decision module

262 adders

264 accumulators

270 decision unit

272 comparators

274 holders

Embodiment

See also Fig. 1, Fig. 1 is the embodiment flow chart of method proposed by the invention.Method proposed by the invention can be used in the video compression system, and to carry out moltion estimation, this video compression system is divided into a plurality of blocks with a present picture.Each step among the figure one below will be described in detail in detail:

Step 110: for the current encoder block in this present picture, (x, y), wherein this current encoder block includes (i1-i0) * (j1-j0) individual pixel progressively to inspect the interior a plurality of candidate motion vectors of a Search Area.

Step 120: (x y), calculates following cost function for each candidate motion vector

$\mathrm{CF}(x,y)=\underset{i=i0}{\overset{i1}{\mathrm{\Σ}}}\underset{j=j0}{\overset{j1}{\mathrm{\Σ}}}\left\{\right|C_{i,j}-P_{i+x,j+y}|+h(\mathrm{Qp},C_{i,j},P_{i+x,j+y})\}$

C wherein _{I, j}Be the pixel in this current encoder block, P _{I+x, j+y}Be the pixel in this Search Area in the target picture, Qp then is a quantization step (quantization step).

Step 130: with can make in this Search Area this cost function draw minimum value candidate motion vector (x, y) be made as this current encoder block the suitableeest motion vector (x1, y1).

The employed method of method of the present invention and known technology has the difference of a maximum, be exactly the employed cost function of the inventive method not only considered the antipode summation (promptly | C _{I, j}-P _{I+x, j+y}|), also considered that the item that is relevant to quantization step Qp and antipode summation distribution scenario (is h (Qp, C _{I, j}, P _{I+x, j+y})).

For instance, function h (Qp, C _{I, j}, P _{I+x, j+y}) can be expressed as K * f (Qp) * g (| C _{I, j}-P _{I+x, j+y}|), and K is a constant, function f (Qp) then is a monotone increasing function (monotonic increasingfunction), promptly when the value of Qp was big more, the value of function f (Qp) also will be big more, this moment h (Qp, C _{I, j}, P _{I+x, j+y}) will be important more for the influence of cost function value.As for function g (| C _{I, j}-P _{I+x, j+y}|), then be used for roughly calculating the current encoder block with respect to the candidate motion vector of this moment (x, the y) changing condition of the pixel difference value of the block that is corresponded to (distribution that has promptly reflected the residual value that is positioned at frequency field), when $\underset{i=i0}{\overset{i1}{\mathrm{\Σ}}}\underset{j=j0}{\overset{j1}{\mathrm{\Σ}}}g\left(\right|C_{i,j}-P_{i+x,j+y}\left|\right)$ Value more little, usually (x y) can cause the residual value that calculates can reach compression efficiency preferably afterwards via the computing that the cosine conversion of loosing, quantification program, saw chi scan (or other the mode that scans) and variable length code or the like just to have represented this candidate motion vector of use.

With next enumerate function g (| C _{I, j}-P _{I+x, j+y}|) an example:

As for ad_max and ad_min then shown in following formula:

In above-mentioned example, whenever one group (i, j) value correspond to function g (| C _{I, j}-P _{I+x, j+y}|) value calculated after, the value of ad_max and ad_min can be upgraded, with calculated for subsequent (i, combination j) correspond to function g (| C _{I, j}-P _{I+x, j+y}|) value the time use.

With size is that 2 * 3 block is an example, and (i1-i0)=2, (j1-j0)=3 if having only two candidate motion vectors in this Search Area, correspond to one first block respectively and unify second block at this moment; This current encoder block with respect to the absolute value of the first block of pixel difference be 8,9,6,8,7,6}, this current encoder block with respect to the absolute value of the second block of pixel difference be 5,10,4,22,0,0}.Use known technology with the computational methods of antipode summation as cost function, the cost function value that first block and second block are calculated can be respectively:

First block: 8+9+6+8+7+6=44

Second block: 5+10+4+22=41

So the time known technology can be with the most identical block of second block as this current encoder block.

Yet if use the computational methods of the foregoing description, suppose K=1, f (Qp)=1, the cost function value that first block and second block are calculated then can be respectively:

First block: [8+0]+[9+ (9-8)]+[6+ (8-6)]+[8+0]+[7+0]+[6+0]=47

Second block:

[5+0]+[10+(10-5)]+[4+(5-4)]+[22+(22-10)]+[0+(4-0)]+[0+0]＝63

Clearly, use the account form that the foregoing description proposed, first block can be selected as the most identical block of this current encoder block, and this point has significantly different with known technology.

And if the use function g that the foregoing description proposed (| C _{I, j}-P _{I+x, j+y}|), cost function then proposed by the invention can abbreviation be:

$\mathrm{CF}(x,y)=\underset{i=i0}{\overset{i1}{\mathrm{\Σ}}}\underset{j=j0}{\overset{j1}{\mathrm{\Σ}}}|C_{i,j}-P_{i+x,j+y}|+K\·f\left(\mathrm{Qp}\right)\·(\mathrm{ad}\_\max -\mathrm{ad}\_\min )$

Wherein ad_max and ad_min be respectively from (i, j)=(i0, j0) to (i, j)=(i1, j1) between | C _{I, j}-P _{I+x, j+y}| maximum and minimum value.

That is to say, use method of the present invention, an if candidate motion vector (x in this target picture, y) block that is corresponded to is with respect to the distribution of the pixel difference value of this current encoder block big more (difference that is in above-mentioned example between maximum and minimum value is big more), the resulting value of function that then assesses the cost also will be big more, and such block just is not easy selected the most identical block as this current encoder block.

See also Fig. 2, Fig. 2 is used for the schematic diagram that a video compression system carries out the device 200 of moltion estimation for the present invention.This video compression system is divided into a plurality of blocks with a present picture, for the current encoder block in this present picture, device 200 can be in a Search Area a plurality of candidate motion vector (x, y) determine a suitableeest motion vector (x1 in, y1), wherein include (i1-i0) * (j1-j0) individual pixel in this current encoder block.In Fig. 2, device 200 includes: an antipode computing unit 220 is used for calculating | C _{I, j}-P _{I+x, j+y}| value, C wherein _{I, j}Be the pixel in this current encoder block, P _{I+x, j+y}It is the pixel in this Search Area in the target picture; One spatial variations computing module 240 is used for calculating function h (Qp, a C _{I, j}, P _{I+x, j+y}) value, wherein Qp is a quantization step; And a motion vector decision module 260, be coupled in antipode computing unit 220 and spatial variations computing module 240, be used for calculating following cost function

$\mathrm{CF}(x,y)=\underset{i=i0}{\overset{i1}{\mathrm{\Σ}}}\underset{j=j0}{\overset{j1}{\mathrm{\Σ}}}\left\{\right|C_{i,j}-P_{i+x,j+y}|+h(\mathrm{Qp},C_{i,j},P_{i+x,j+y})\}$

And with in this Search Area make this cost function draw minimum value candidate motion vector (x, y) be set at this current encoder block the suitableeest motion vector (x1, y1).

For instance, as described before, function h (Qp, C _{I, j}, P _{I+x, j+y}) can be expressed as K * f (Qp) * g (| C _{I, j}-P _{I+x, j+y}|), so in the device 200 in, include one first computing unit 242 in the spatial variations computing module 240, be used for calculating function g (| C _{I, j}-P _{I+x, j+y}|) value; One second computing unit 244 is used for calculating the value of function f (Qp); And a multiplier 246, be coupled in first computing unit 242 and second computing unit 244, be used for calculating K * f (Qp) * g (| C _{I, j}-P _{I+x, j+y}|) value.

Include as for 260 of motion vector decision module: an adder 262, be coupled in antipode computing unit 220 and spatial variations computing module 240, be used for calculating | C _{I, j}-P _{I+x, j+y}|+h (Qp, C _{I, j}, P _{I+x, j+y}) value; One accumulator 264 is coupled in adder 262, is used for out the value of this cost function of calculating; And one the decision unit 270, be coupled in accumulator 246, be used for can make in this Search Area this cost function draw minimum value candidate motion vector (x, y) be set at this current encoder block the suitableeest motion vector (x1, y1).

And in order to make decision unit 270 have the function of the size of the function of storage cost function minimum and the different cost function values that candidate motion vector calculates of comparison, in Fig. 2,270 of unit of decision include a comparator 272, be coupled in accumulator 264, be used for output the suitableeest motion vector (x1, y1); And a holder 274, be coupled in 272 comparators, be used for storing the minimum value that this cost function occurred.

Use method and relevant apparatus proposed by the invention, can calculate the suitableeest motion vector of a current encoder block in a Search Area according to the cost function that is different from known technology.Because cost function used in the present invention can be subjected to this current encoder block with respect to a candidate motion vector (x, y) changing condition of the value differences of the block that is corresponded to influences, so use method proposed by the invention and relevant apparatus can guarantee that the corresponding residual value of the suitableeest motion vector that calculates can have compression efficiency preferably afterwards through handling (be above-mentioned diffusing cosine conversion, quantification program, saw that chi scans, the computing of variable length code or the like).

The above only is preferred embodiment of the present invention, and all equalizations of being done according to the present patent application claim change and modify, and all should belong to the covering scope of patent of the present invention.

Claims (9)

1. one kind is used for the method that a video compression system carries out moltion estimation, and this video compression system will

One present picture is divided into a plurality of blocks, and this method includes following steps:

For the current encoder block in this present picture, progressively inspect a plurality of candidate motion vectors in the Search Area (x, y), wherein this current encoder block includes (i1-i0) * (j1-j0) individual pixel;

(x y), calculates following cost function for each candidate motion vector

$\mathrm{CF}(x,y)=\underset{i=i0}{\overset{i1}{\mathrm{\Σ}}}\underset{j=j0}{\overset{j1}{\mathrm{\Σ}}}\left\{\right|C_{i,j}-P_{i+x,j+y}|+h(\mathrm{Qp},C_{i,j},P_{i+x,j+y})\}$

Wherein, C _{I, j}Be the pixel in this current encoder block, P _{I+x, j+y}Be the pixel in this Search Area in the target picture, Qp then is a quantization step; And

With can make in this Search Area this cost function draw minimum value candidate motion vector (x, y) be made as this current encoder block the suitableeest motion vector (x1, y1).

2. the method for claim 1, wherein function h (Qp, C _{I, j}, P _{I+x, j+y}) can be expressed as K * f (Qp) * g (| C _{I, j}-P _{I+x, j+y}|), and K is that a constant, function f (Qp) they are a monotone increasing function.

3. the method for claim 1, wherein if other condition is constant, then when the value of Qp is big more, function h2 (Qp, C _{I, j}, P _{I+x, j+y}) value promptly can be big more.

4. one kind is used for the device that a video compression system carries out moltion estimation, this video compression system is divided into a plurality of blocks with a present picture, for the current encoder block in this present picture, this device can be in a Search Area a plurality of candidate motion vector (x, y) determine a suitableeest motion vector (x1 in, y1), wherein include (i1-i0) * (j1-j0) individual pixel in this current encoder block, this device includes:

One antipode computing unit is used for calculating | C _{I, j}-P _{I+x, j+y}| value, C wherein _{I, j}Be the pixel in this current encoder block, P _{I+x, j+y}It is the pixel in this Search Area in the target picture;

One spatial variations computing module is used for calculating function h (Qp, a C _{I, j}, P _{I+x, j+y}) value, wherein Qp is a quantization step; And

One motion vector decision module is coupled in this antipode computing unit and this spatial variations computing module, is used for calculating following cost function

$\mathrm{CF}(x,y)=\underset{i=i0}{\overset{i1}{\mathrm{\Σ}}}\underset{j=j0}{\overset{j1}{\mathrm{\Σ}}}\left\{\right|C_{i,j}-P_{i+x,j+y}|+h(\mathrm{Qp},C_{i,j},P_{i+x,j+y})\}$

And with in this Search Area make this cost function draw minimum value candidate motion vector (x, y) be set at this current encoder block the suitableeest motion vector (x1, y1).

5. device as claimed in claim 4, wherein, function h (Qp, C _{I, j}, P _{I+x, j+y}) can be expressed as K * f (Qp) * g (| C _{I, j}-P _{I+x, j+y}|), and K is that a constant, function f (Qp) they are a monotone increasing function.

6. device as claimed in claim 5, wherein, this spatial variations computing module includes:

One first computing unit, be used for calculating function g (| C _{I, j}-P _{I+x, j+y}|) value;

One second computing unit is used for calculating the value of function f (Qp); And

One multiplier is coupled in this first computing unit and this second computing unit, be used for calculating K * f (Qp) * g (| C _{I, j}-P _{I+x, j+y}|) value.

7. device as claimed in claim 4, wherein, this motion vector decision module includes:

One adder is coupled in this antipode computing unit and this spatial variations computing module, is used for calculating | C _{I, j}-P _{I+x, j+y}|+h (Qp, C _{I, j}, P _{I+x, j+y}) value;

One accumulator is coupled in this adder, is used for out the value of this cost function of calculating; And

One the decision unit, be coupled in this accumulator, be used for can make in this Search Area this cost function draw minimum value candidate motion vector (x, y) be set at this current encoder block the suitableeest motion vector (x1, y1).

8. device as claimed in claim 7, wherein, this decision unit pack contains:

One comparator is coupled in this accumulator, be used for exporting this suitableeest motion vector (x1, y1); And

One holder is coupled in this comparator, is used for storing the minimum value that this cost function occurred.

9. device as claimed in claim 4, wherein, if other condition is constant, then when the value of Qp is big more, function h (Qp, C _{I, j}, P _{I+x, j+y}) value promptly can be big more.