CN1163540A - Method and device for deducing fine movement - Google Patents

Abstract

A method for motion estimation includes steps for constructing a first image segment representing a portion of present image frame and a second image segment having a plurality of image segments of an adjacent frame, comparing said plurality of image segments of said second image segment with the image signal within the first image segment, and generating a number of error values, detecting a first motion vector at one scale of resolution using the generated error values, outputting the error value associated with a reference segment pointed by the first motion vector, and the error values corresponding to the reference segment shifted by one unit of selected resolution respectively in selected direction(s), generating a second motion vector in finer scale of resolution using the output error values, and summing the first and second motion vectors.

Description

The method and apparatus of deducing fine movement

The present invention relates to the method and apparatus of deducing fine movement, relate in particular to method and apparatus motion vector, that be used under the motion compensation DPCM of coded digital picture signal mode, inferring the very fine motion of further inferring the precise image resolution scale with the error amount that method produced of the motion vector of inferring the low image resolution scale.

In order more effectively to compress the data that transmit,, generally use diversified coding method using in the image signal processing apparatus of continuous number picture signal as HDTV (high-definition television), digital VTR (tape record machine) and multimedia etc.Wherein, DPCM (digital pulse-code modulation) mode is to utilize the data dependence that exists between the picture signal consecutive frame to come data image signal is encoded.

With the difference signal coding of DPCM mode adjacent interframe on to the time time, the many data volume that the field produced of interframe movement is more than the field that motion is lacked.But, if by finding out shift position, being the shift position the similar field of the image of present frame specific area and consecutive frame from contiguous each fragment (segment) to the specific fragment of present frame, come differential coding to two fields, then data volume has reduced, and thinks that this mode is the DPCM mode through motion compensation.Therefore, in the DPCM mode after motion compensation, encode, then can improve transmission efficiency by image segments that present frame has been set and the difference data between the consecutive frame correspondence image fragment.According to motion deduction mode, motion vector is represented a kind of like this direction of motion and motion size of image segments, this image segments be around the image segments of exploring present frame and the consecutive frame correspondence image fragment time between two frames difference signal be minimum image segments.

Deduction about the pixel unit motion has proposed multiple mode.In general, because that the time is gone up the pixel unit of the motion between two different frames is littler than integer pixel unit, carry out the data compression that motion compensation does not often accomplish that desirable efficient is higher with the motion vector of integer pixel unit.Therefore, if use the motion vector of sub-pixel unit, little when then the difference of interframe compares the motion vector that uses integer pixel unit.One of sub-pixel unit is a half-pix unit, in order to infer motion, uses the deduction method of half-pix unit usually.Illustrating briefly below according to existing mode uses half-pix unit to infer the method for motion.

The device of Fig. 1 comprises: receive current frame image signal, and the image segments that forms the image segments that constitutes by a plurality of pixels form device 1; Receive the consecutive frame picture signal, and the exploration field that forms the image segments that constitutes by a plurality of pixels form device 2; Reception from image segments form device 1 and exploration field form the picture signal of device 2 outputs and detect pixel unit the first motion vector MV1 first explore unit 3; Reception forms device 2 and first and explores the picture signal of unit 3 outputs, and uses the half-pix interpolation unit 4 that is calculated the image signal value on the half-pixel position of benchmark pixel periphery by first pixel of exploring image segments that becomes benchmark fragment in the consecutive frame that the unit 3 detected first motion vector MV1 are determined and periphery thereof from the exploration field; Reception from image segments form the picture signal of device 1 and 4 outputs of half-pix interpolation unit and export half-pix unit the second motion vector MV2 second explore unit 5, and receive and to explore unit 3 and second from first and explore the signal MV1 that unit 5 export respectively, MV2 and their phases adder unit 6 of output movement vector M V in addition.

Image segments forms device 1 and receives the picture signal of present frame and store input signal with the form of fragment.Along with the input of current frame image signal, the exploration field forms the picture signal around the respective segments that device 2 receives consecutive frames and forms exploration field fragment.The most comparable fragment of fragment with present frame is explored in the first exploration unit 3 in the fragment of the field of exploration.By comparing the data of the pixel unit between two fragments, will output to adder unit 6 and half-pix interpolation unit 4 from the first first motion vector MV1 that explores the integer pixel unit of unit 3 generations.The value that half-pix interpolation unit 4 will calculate carry out linear interpolation with the benchmark pixel of the fragment of the first motion vector MV1 appointment of integer pixel unit and neighboring pixel thereof in consecutive frame outputs to second and explores unit 5.

According to above-mentioned existing systems, image segments has by N ₁* N ₂The shape of the piece that pixel constitutes, the exploration field is by M ₁* M ₂Pixel constitutes.As another kind of state, image segments and exploration field can be according in the image shape and size as body being formed.In addition, as the P frame of MPEG (dynamic image compression international standard) or only with the occasion of only in reference frame, asking the single directional prediction of a consecutive frame the B frame of backward-predicted, the image segments in the exploration field that constitutes in the image segments of present frame and frame or the future frame is in the past compared.Yet above-mentioned motion is inferred and also is applicable to bi-directional predicted with past frame and future frame both direction.In bi-directional predicted occasion, the image segments in the exploration field that forms in the image segments of present frame and frame in the past and the future frame is compared as the MPEGB frame.

Another the existing method that is used to form image segments and explores the field be accomplish with image in consistent to the picture body.Therefore, the image segments of formation and the field of exploration fragment can have any size and shape.

Among Fig. 2, the interval of " 0 " remarked pixel unit, the interval of " X " expression half-pix unit.Second explores image segments that unit 5 makes the consecutive frame that the first motion vector MV1 according to integer pixel unit forms in that each moves half-pixel unit, explores the position, the output vector component (1/2 that have the minimum movement compensating error among the position (X) with respect to eight half-pixel motion vector of gained and the first motion vector MV1 again on all directions, 0,1/2) one of, promptly with the numerical value of the meticulous control of half-pixel unit.First motion vector MV1 of the integer pixel unit that adder unit 6 will be exported from the first exploration unit 3 and the second motion vector MV2 addition of the half-pix unit that exports from the second exploration unit, thereby the MV of motion vector completely of generation do motion deduction usefulness.For example, adder unit 6 by will from second explore unit 5 half-pix unit the second motion vector MV2 horizontal component (1/2) with from first motion vector the MV1 (=x that explores the pixel unit of unit 3, y) addition decide motion vector MV (=x-1/2/, y).

In the occasion of inferring with the bidirectional-movement of the past and the consecutive frame in future, same procedural application is in forward direction and back to two kinds of predictions.

Above-mentioned existing method is done linear interpolation with the neighboring pixel of determined fragment and this fragment to the respective pixel value on half-pixel position then with the position of pixel separation unit's decision fragment.This method is used by the fragment pixel value of the present frame of the motion vector decision of the half-pix value of interpolation and integer pixel unit and is selected the most similar fragment among eight half-pix movement positions (being Fig. 2 " X ").Therefore, the shortcoming of so existing method is to be used to infer that the processing of motion vector of half-pix unit is extremely huge.This shortcoming is not limited in the occasion that half-pix unit's motion vector is inferred, for inferring the occasion of motion vector with the image resolution ratio meticulousr than pixel unit, this also is the general problem that exists.

The present invention proposes for addressing the above problem, its objective is for the effectively compensating that the motion that makes very fine is provided and become possible motion deduction method, promptly provide a kind of motion deduction method, the motion vector that it uses a kind of error amount of image resolution ratio scale to infer trickleer image resolution ratio scale, image segments of having set in the comparison present frame for the motion vector that detects a kind of image resolution ratio scale and the data between the image segments in the exploration field, compare the error amount that produces by above-mentioned data comparative result again and the motion vector of inferring precise image resolution.

Another object of the present invention provides a kind of motion apparatus for predicting, this device produces a kind of image resolution ratio scale that is used to infer the interframe movement vector according to the data comparison between image segments of having set in the present image and the consecutive frame, and adopts error amount that obtains according to this image resolution ratio scale and the error amount that is relatively produced to make effective compensation of the very trickle motion of the motion vector of inferring meticulous scale become possibility.

In order to achieve the above object, fine movement deduction method of the present invention comprises: form the expression current image frame a part first image segments and have the step of second image segments of a plurality of image segments of consecutive frame; The above-mentioned a plurality of image segments and the picture signal in first image segments of above-mentioned second image segments are made comparisons to produce the step of a plurality of error amounts; Detect a kind of step of first motion vector of image resolution ratio scale with the error amount that is produced; Output with by the relevant error amount of the benchmark fragment of the first motion vector indication and output step corresponding to the error amount of the benchmark fragment after moving, wherein each moves the selected digital image resolution scale of 1 unit to the benchmark fragment of Yi Donging in the direction of selecting separately; Produce the step of second motion vector of precise image resolution scale with the error amount of output; And the step of first motion vector and the second motion vector addition.

Again, fine movement apparatus for predicting of the present invention, be picture signal between a kind of relatively consecutive frame to infer the device of image motion, comprising: receive the picture signal of present frame and form the equipment of first image segments; Receive the picture signal of consecutive frame and form the equipment of second image segments; Receive the picture signal that forms device and the output of second image segments formation device respectively from first image segments, detect first motion vector of selected a kind of image resolution ratio scale unit, determine the benchmark image of above-mentioned second image segments consistent with above-mentioned first motion vector, each moves the selected digital image resolution scale of 1 unit and will move the peripheral image segments that produced thus and the picture signal of the benchmark image fragment and first image is made comparisons towards selected direction with the benchmark image fragment, and the exploration equipment of the error amount output that will draw by each comparative result; Reception is by the error amount of exploration equipment output and produce the control appliance of second motion vector of the image resolution ratio scale meticulousr than above-mentioned selected scale; And receive first and second motion vectors and with the addition equipment of their additions and output.

Describe the preferred embodiments of the present invention in detail below in conjunction with accompanying drawing.

Fig. 1 is the block diagram that expression is used to infer the existing apparatus of half-pix motion;

Fig. 2 is the schematic diagram of remarked pixel unit and half-pix unit;

Fig. 3 A is the block diagram of the motion apparatus for predicting of expression one embodiment of the present invention, and Fig. 3 B is the block diagram of the motion apparatus for predicting of another preferred embodiment of expression the present invention;

Fig. 4 is the detailed diagram of the horizontal half-pix controller of presentation graphs 3A and Fig. 3 B;

Fig. 5 A and 5B are the schematic diagrames of the half-pixel motion vector determining method of expression horizontal half-pix controller A and B.

Though detailed description of the present invention is the selection of making in order to make the meticulous scale be used for the unit scale of first motion vector and be used for second motion vector become pixel unit image resolution ratio and half-pix unit's image resolution ratio respectively, but this example does not limit the present invention, and this detailed description also is applicable to the scale of different choice.

According to a fragment in the consecutive frame, the motion vector of the fragment of having set in the present frame show as MV=(x, y).Wherein motion vector MV has the horizontal component x and the vertical component y of motion vector.

Fig. 3 A is the block diagram of the motion apparatus for predicting of expression one embodiment of the present invention.

The device of Fig. 3 A comprises: receive present frame picture signal, and the image segments that forms the image segments that constitutes by a plurality of pixels form device 31; Receive consecutive frame picture signal, and the exploration field that forms the image segments that constitutes by a plurality of pixels form device 32; Reception forms the picture signal of device 31 and 32 outputs of exploration field formation device from image segments, according to form the first motion vector MV1 that image segments in the exploration field fragment that device 32 forms and the picture signal comparative result between the image segments in the present frame detect pixel unit by the exploration field, according to the pointed image segments that is called the benchmark fragment of the detected first motion vector MV1 with according to each moves image segments that a pixel produces and the exploration unit 33 of output error value in level and vertical direction with this benchmark fragment; Reception from the error amount of exploring unit 33 outputs, and vertically and flatly calculate the half-pix control unit 34 of the half-pix second motion vector MV2 at interval respectively; Reception is from the first motion vector MV1 that explores unit 33 and from the second motion vector MV2 of half-pix control unit 34 and export the adder unit 37 of the motion vector as a result of each addition of vectors.

Half-pix control unit 34 has the vertical half-pix controller 36 that the horizontal component horizontal half-pix controller of using 35 and the vertical component that detects the second motion vector MV2 that detect the second motion vector MV2 used.

Among the picture signal with two consecutive frames of timing separation, image segments forms the picture signal that device 31 receives present frames and also stores input signal in first fragment.In company with the input of current frame image signal, the exploration field forms device 32 and receives the picture signal of consecutive frame, the picture signal that also preferably storage is received in second fragment bigger than first fragment.Exploring unit 33 individually receives respectively from exploration field formation device 32 and second fragment of having stored of image segments formation device 31 outputs and the picture signal of first fragment.Under the occasion of P frame, between present frame and past frame, compare.And under the occasion of B frame, except that comparing between present frame and the past frame, also between present frame and future frame, comparing.So under the occasion of B frame, the exploration field forms device 32 and forms second fragment of past frame and the 3rd fragment of future frame.Thereby compare operation is to carry out between present frame and past frame earlier, carries out between present frame and future frame then, or in contrast, perhaps compares simultaneously shown in Fig. 3 B.

Illustrate then that below first fragment and second fragment have N respectively ₁* N ₂Pixel and M ₁* M ₂The occasion of the shape of pixel.And for the simplification that illustrates, selecting integer pixel unit is a kind of image resolution ratio scale, and selecting half-pix unit is the precise image resolution scale.Certainly, the present invention also can be applicable to the image resolution ratio scale of different choice.

The motion inference error that is obtained by the motion vector of inferring integer pixel unit is expressed as P ₀, by each moves a pixel and the motion inference error that produces is expressed as P respectively in level and vertical direction on the basis of the motion vector of inferring ₁', P ' _-1, P _-1And P ₁Mean absolute error (MAE) or average two takes advantage of error (MSE) can usefully be used in the calculating of motion inference error.Come calculated level motion inference error P by MAE ₀, P ₁, P _-1, the preferably following formula of spendable for this reason mathematical formulae. ${\mathrm{<figure-callout\; id="0"\; label="P"\; filenames="A9710223700181.png"\; state="\{\{state\}\}">P</figure-callout>}}_0=\underset{m=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{n=0}{\overset{N-1}{\mathrm{\&Sigma;}}}|Y(\mathrm{Nk}+m).\mathrm{Nl}+n)-$

Y′(Nk+m+x、Nl+n+y)｜ $P_1=\underset{m=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{n=0}{\overset{N-1}{\mathrm{\&Sigma;}}}|Y(\mathrm{Nk}+m).\mathrm{Nl}+n)-$

Y′(Nk+m+x+1、Nl+n-y)｜ $P_{-1}=\underset{m=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{n=0}{\overset{N-1}{\mathrm{\&Sigma;}}}|Y(\mathrm{Nk}+m).\mathrm{Nl}+n)-$

Y′(Nk+m+x-1、Nl+n+y｜

Wherein, Y represents the picture signal of present frame, the picture signal of Y ' expression consecutive frame, N ₁=N ₂=N represents the size of fragment.In above-mentioned numerical expression, (k, l) expression is from (k, position l) of the upper right quarter of frame.Inference error P ' moves both vertically ₁, P ' _-1Available above-listed formula calculates with same method.

Explore unit 33 and the image segments of present frame and a plurality of adjacent image fragments with similar size in consecutive frame exploration field are made comparisons, and calculate the motion inference error of indivedual comparative results with pixel unit.The first motion vector MV1 of integer pixel unit is decided by the position of motion inference error for minimum fragment, and outputs to adder unit 7.Benchmark image fragment in the consecutive frame that exploration unit 33 will be determined according to the first motion vector MV1 of integer pixel unit and the motion inference error P that obtains according to a plurality of image segments that move a pixel position on every side each of benchmark image fragment _-1, P ₁, P ₀, P ' _-1, P ' ₁Output to half-pix control unit 34.Be more preferably, the present invention can determine the level and the vertical component of the motion vector of half-pix unit simultaneously independently.And since horizontal half-pix controller 35 and vertical half-pix controller 36 make in the same way, so the method for the horizontal component of explanation decision half-pixel motion vector only.

The occasion of unidirectional motion prediction, the motion inference error P of the comparative result of expression present frame and consecutive frame _-1, P ₁, P ₀, P ' _-1, P ' ₁Output to pixel control unit 34 from exploring unit 33 as previously mentioned.Yet for the prediction of the bidirectional-movement as the B frame, the error that is obtained by present frame and past frame comparative result outputs to half-pix control unit 34 with the motion inference error that is obtained by present frame and future frame comparative result from exploring unit 33.

Fig. 3 B, except two consecutive frames are imported this system simultaneously, similar to Fig. 3 A.Under the occasion of B frame, future frame and past frame are imported this system simultaneously.

Fig. 4 is the detailed diagram of the horizontal half-pix controller of presentation graphs 3A or Fig. 3 B.

The machine configuration of Fig. 4 has: be used to receive motion inference error P ₁, P ₀, P _-1Three inputs 41,42,43; Input signal P with first input end 41 ₁Input signal P with second input 42 ₀The first adder A1 of phase adduction output; Input signal P with second input 42 ₀Input signal P with the 3rd input 43 _-1The second adder A2 of phase adduction output; Have input of having set that is connected with first adder A1 output and the first comparator C MP1 that compares each input signal; Have input of having set that is connected with second adder A2 output and the second comparator C MP2 that compares each input signal; Multiply by coefficient of having set and the first multiplier M1 that its product value is outputed to the first comparator C MP1 for second adder A2 output signal; Multiply by coefficient of having set and the second multiplier M2 that its product value is outputed to the second comparator C MP2 for first adder A1 output signal; And to each output signal of coming from comparator C MP1, CMP2 do negative logic and NOR gate NOR.

From exploring the motion inference error P of unit 33 outputs ₁, P ₀, P _-1Be input to the horizontal half-pix controller 35 of half-pix control unit 34.So first adder A1 is from the input signal P by first input end 41 ₁Deduct input signal P by second input 42 ₀Second adder A2 is from the input signal P by the 3rd input 43 _-1Deduct input signal P by second input 42 ₀The first and second multiplier M1, M2 multiply by adder output signal each coefficient of having set respectively and the result of multiplying are outputed to comparator C MP1, the CMP2 that is connected respectively to the multiplier output.More best is P ₀And P _-1Difference a and P ₀And P ₁Difference b individually be input to the second comparator C MP2 and the first comparator C MP1.Be more preferably, the first comparator C MP1 compares the output signal of the first adder A1 and the first multiplier M1 and exports comparative result.The second comparator C MP2 compares the output signal of the first adder A2 and the second multiplier M2 and exports comparative result.

Fig. 5 A and Fig. 5 B are the schematic diagrames in the method for the motion vector of horizontal half-pix controller 35 decision half-pix units.Fig. 5 A represents the left of the motion vector MV2 of half-pix unit at the motion vector MV1 of pixel unit, and the motion vector MV2 that Fig. 5 B represents half-pix unit is right-hand the motion vector MV1's of pixel unit.

At Fig. 5 A, P ₁Compare P _-1(that is, b is bigger than a) greatly, it is last that this true expression motion vector MV is positioned at the distance of left about 1/2 of the first motion vector MV1 of pixel unit.At Fig. 5 B, with P ₀Be benchmark, P _-1Compare P ₁(that is, a is bigger than b) greatly, it is last that this true expression motion vector MV is positioned at right-hand about 1/2 the distance of the first motion vector MV1 of pixel unit.

At Fig. 5 A, for P ₀Be benchmark P ₁Compare P _-1Big occasion has only the output 44 of comparator C MP1 to be " height ".Therefore, horizontal half-pix controller 35 outputs " 1/2 " are as the horizontal component value of the second motion vector MV2.

At Fig. 5 B, for P ₀Be benchmark P _-1Compare P ₁Big occasion has only the output 46 of comparator C MP2 to be " height ".Therefore, horizontal half-pix controller 35 outputs " 1/2 " are as the horizontal component value of the second motion vector MV2.With P ₀Be benchmark P ₁And P _-1Under the identical mutually occasion, the output signal both sides of two comparator C MP1, CMP2 are " low ".Therefore, NOR gate NOR receives " low " value and output " height " signal.Therefore, the horizontal component of the second motion vector MV2 is " 0 ".

Adder unit 37 receives the second motion vector MV2, with its first motion vector MV1 that is added to the pixel cell that is obtained by exploration unit 33, infers the motion vector MV of half-pix unit thus.

As mentioned above, motion deduction method of the present invention and device, in order under selected image resolution ratio scale, to detect motion vector, relatively the picture signal of fragment in selected fragment and the consecutive frame in the present frame, and produce according to selected motion vector and obtain error amount in all directions calculating of this motion vector.

By the error amount that vertical comparison produced, detect the vertical component of the motion vector of precise image resolution scale.By the error amount that the level comparison is produced, detect the horizontal component of the motion vector of precise image resolution scale.Therefore, owing to reduced the data volume that should handle of the motion vector that is used to detect the precise image resolution scale, the present invention has improved the efficient that motion is inferred significantly.

The present invention illustrates with image segments and exploration field with specific modality, still needs only image segments and exploration field and is formed by shape and size to the picture body, and this image segments just can form any shape and size with the plain field of spy.In addition, the fragment of present frame more not only with the time on the past frame or the future frame that leave more than present frame one frame carry out, also can be right after before it or carry out at thereafter frame.Furtherly, compare operation can also carried out in conjunction with the arbitrary frame in the above-mentioned frame as the B frame in the mpeg system of more more than one consecutive frame.

Claims (25)

1, the picture signal between the consecutive frame and infer the method for image motion relatively is characterized in that comprising the steps:

Form the step of second image segments of first image segments of a part of representing current image frame and a plurality of image segments that formation has consecutive frame;

Picture signal in the described a plurality of image segments of more described second image segments and first image segments also produces the step of a plurality of error amounts;

Detect a kind of step of first motion vector of image resolution ratio scale with the error amount that is produced;

According to the connected error amount of benchmark fragment of the first motion vector indication with according to each moves the step that benchmark fragment after the moving of obtaining after the selected digital image resolution of a unit is come the output error value with preferential direction;

Produce the step of second motion vector of precise image resolution scale with the error amount of output; And

Step with first motion vector and the second motion vector addition.

2, the method for claim 1 is characterized in that described generation step comprises:

With the error amount of described benchmark image fragment and place error amount with respect to the described peripheral image segments of the first direction at described benchmark image fragment center to detect the step to the component of described first direction of described second motion vector;

With the error amount of described benchmark image fragment and place error amount with respect to the described peripheral image segments of the second direction at described benchmark image fragment center to detect the step to the component of described second direction of described second motion vector.

3, method as claimed in claim 2 is characterized in that described first direction and second direction are respectively horizontal direction and vertical direction.

4, method as claimed in claim 2 is characterized in that to the described component of described second motion vector of described first direction and the position of satisfying the image segments of minimum error values to the described representation in components of described second motion vector of described second direction.

5, the method for claim 1 is characterized in that described consecutive frame is a past frame.

6, method as claimed in claim 5 is characterized in that described past frame is the frame that was right after before described present frame.

7, method as claimed in claim 5 is characterized in that described past frame is a plurality of frames that make a distinction in time from described present frame.

8, the method for claim 1 is characterized in that described consecutive frame is a future frame.

9, method as claimed in claim 8 is characterized in that described future frame is the frame that is right after after described present frame.

10, method as claimed in claim 8 is characterized in that described future frame is a plurality of frames that make a distinction in time from described present frame.

11, the method for claim 1 is characterized in that described a plurality of image segments of described second image segments have respectively and the same clip size of described first image segments.

12, the method for claim 1 is characterized in that described a kind of image resolution ratio scale is a pixel unit, and described precise image resolution scale is a half-pix unit.

13, the picture signal between the consecutive frame and infer the device of image motion relatively comprises:

Receive current frame image signal and form the equipment of first image segments;

Receive the consecutive frame picture signal and form the equipment of second image segments;

Receive the picture signal that forms device and the output of second image segments formation device respectively from first image segments, detect first motion vector of selected a kind of image resolution scale unit, determine the benchmark image of described second image segments consistent with described first motion vector, by each selected digital image resolution scale that moves a unit produces peripheral image and the peripheral image segments that produced and the picture signal of benchmark image fragment are compared with the picture signal of first image segments with selected direction with the benchmark image fragment, the exploration equipment of the error amount that output obtains from each result relatively;

Reception is from the error amount of exploration equipment output and produce the control appliance of second motion vector of precise image resolution scale according to described selected scale; And

Receive first and second motion vector and with the addition equipment of its addition and output.

14, device as claimed in claim 13 is characterized in that described control appliance comprises:

The described error amount of more described benchmark image fragment with place with respect to the error amount of the described peripheral image segments of the first direction at described benchmark image fragment center and detect first controller to the component of a vector of described first direction of described second motion vector; And

The described error amount of more described benchmark image fragment with place with respect to the error amount of the described peripheral image segments of the second direction at described benchmark image fragment center and detect second controller to the component of a vector of described second direction of described second motion vector.

15, device as claimed in claim 14 is characterized in that described first direction and second direction are respectively horizontal direction and vertical direction.

16, device as claimed in claim 14 is characterized in that described controller comprises:

From deduct first subtracter of the error amount that gets according to described benchmark image fragment according to the error amount that is positioned at described benchmark image fragment one side's described peripheral image segments gained;

From deduct second subtracter of the error amount that gets according to described benchmark image fragment according to the error amount that is positioned at described benchmark image fragment the opposing party's described peripheral image segments gained;

Each first output signal of described second subtracter be multiply by the coefficient set to generate first multiplier of first product value;

Each second output signal of described first subtracter be multiply by the coefficient set to generate second multiplier of second product value;

The size of each input that receive described second output signal and first product value, relatively receives also provides first comparator of first bianry signal according to comparative result;

The size of each input that receive described first output signal and second product value, relatively receives also provides second comparator of second bianry signal according to comparative result;

Each ground receives described first and second bianry signals of coming from described first and second comparators, carry out negative logic and computing, and generate the NOR gate of the computing income value of described second motion vector of expression.

17, device as claimed in claim 13 is characterized in that described consecutive frame is a past frame.

18, device as claimed in claim 17 is characterized in that described past frame is the frame that was right after before described present frame.

19, device as claimed in claim 17 is characterized in that described past frame is a plurality of frames that make a distinction in time from described present frame.

20, device as claimed in claim 13 is characterized in that described consecutive frame is a future frame.

21, device as claimed in claim 20 is characterized in that described future frame is the frame that is right after after described present frame.

22, device as claimed in claim 20 is characterized in that described future frame is a plurality of frames that make a distinction in time from described present frame.

23, device as claimed in claim 14 is characterized in that described second image segments has a plurality of image segments.

24, device as claimed in claim 23 is characterized in that described a plurality of image segments of described second image segments have respectively and the same clip size of described first image segments.

25, device as claimed in claim 13 is characterized in that described a kind of image resolution ratio scale is a pixel unit, and described precise image resolution scale is a half-pix unit.

Images