CN100421455C - A method for video electronic anti-shake - Google Patents

Abstract

The invention discloses a simple, effective and cost-saving method for video electronic anti-shake. Firstly, the global relative motion vector of the current frame is obtained by using the motion estimation method and the prediction method respectively; secondly, the two obtained by the two methods are calculated. The difference of the global relative motion vector of the current frame, and judge whether there is shaking according to the difference, when there is shaking, use the difference as the motion vector for compensating shaking, when there is no shaking, make the motion vector for compensating shaking is 0; again, it is judged whether the motion vector for compensating for jitter is within the maximum compensation range determined by the preset maximum compensation motion vector value, and if it exceeds this range, the maximum compensation motion vector value is used as the compensation The shake motion vector; finally, on the total pixel image of the current frame, according to the shake-compensated motion vector and the preset effective pixel image size, the effective image after anti-shake compensation of the current frame is obtained by cropping.

Description

A method for video electronic anti-shake

technical field

The invention relates to a video electronic anti-shake method applied in video collection and output equipment.

technical background

With the advancement and development of science and technology, many new functions have been added to equipment such as digital cameras and digital video cameras. But these handheld devices inevitably cause image shaking due to human factors. This vibration avoidance can be achieved through anti-shake technology. The anti-shake system at this stage is mainly divided into two categories: optical anti-shake and electronic anti-shake. Optical anti-shake is to sense the shake of the camera through the built-in instrument of the lens, and then adjust the position of the lens in the lens to achieve the anti-shake effect. Optical image stabilization technology is commonly used in high-end digital cameras and digital video cameras. Electronic image stabilization is to process images by electronic means to reduce the impact of shaking on imaging. At present, there are three main implementation methods of electronic image stabilization, automatic increase of ISO sensitivity/electronic image compression/BSS mode. The principle of CCD anti-shake is to place the CCD on a bracket that can move up and down, left and right, and first detect whether there is shaking. Since the gyro sensor is used, the detection of shaking is basically the same as that of other companies. Then the sensor detects the direction, speed, and amount of movement of the shaking... The detected signal is processed to calculate the amount of CCD movement that can offset the shaking. Electronic image stabilization is usually used in low-end digital cameras and digital video cameras due to its low cost.

It can be seen that the two technologies of optical image stabilization and electronic image stabilization have their own advantages and disadvantages. Optical anti-shake technology has a good anti-shake effect and can make full use of the acquisition pixels of the CCD, but requires a lens motion compensation device, so the cost is relatively high. Electronic anti-shake does not require a lens motion compensation device, which is cheap and easy to implement, but reduces the utilization rate of the CCD; while the CCD compensation method requires an additional CCD moving device, which also increases the cost to a certain extent. Moreover, whether it is optical anti-shake or electronic anti-shake, a motion sensor-shake compensation gyroscope is generally required to sense the direction and intensity of motion, and then optical and electronic methods are used to compensate.

Contents of the invention

The object of the present invention is to provide a simple video electronic anti-shake method that does not require an external motion sensor for the deficiencies of the prior art.

In order to solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a method for video electronic anti-shake, comprising the following steps:

(1) adopt motion estimation method to calculate the current frame and the global relative motion vector of several frame video images before the current frame;

(2) Predict the global relative motion vector of the current frame according to the global relative motion vectors of several frames before the current frame;

(3) Calculate the difference between the global relative motion vector of the current frame obtained in step (1) and step (2), and judge whether there is jitter according to the difference, and when there is jitter, use the difference as compensation jitter The motion vector of , when there is no jitter, the motion vector for compensating jitter is 0;

(4) Judging whether the motion vector for compensating jitter is within the maximum compensation range determined by the preset maximum compensation motion vector value, if it exceeds this range, use the maximum compensation motion vector value as the compensation for jitter motion vector;

(5) On the total pixel image of the current frame, according to the shake-compensated motion vector and the preset size of the effective pixel image, crop the effective image after the anti-shake compensation of the current frame.

Preferably, several frames of video images before the current frame in the step (1) and step (2) can be effective images of each frame; and the video images of the current frame can be pre-cut, and its pre-cut The specific method can be as follows: first obtain the absolute displacement of the previous effective image relative to the center position of the total pixel image; then obtain the center position of the clipping area according to the absolute displacement value, and then use the preset effective pixel image size to crop The center position of the area determines the clipping area, and finally the total pixel image of the current frame is clipped, so as to obtain the video image of the current frame whose size is the size of the effective pixel image.

Preferably, the step (1) can specifically be:

(10), adopt the method for motion estimation to calculate the relative motion vector of search block in each frame image;

(11), count the frequency of the relative motion vector magnitude of each search block, obtain the relative motion vector value with the highest frequency, and calculate the variance of the relative motion vector;

(12), judging whether the relative motion vector value with the highest frequency exceeds the product of the variance of the relative motion vector of the search block in the image and a given threshold, if not exceeded, it indicates that the global relative motion vector of the current frame image is not Obviously, set the global relative motion vector of the current frame to 0; otherwise, use the relative motion vector value with the highest frequency as the global relative motion vector of the current frame.

Preferably, the search block may be all macroblocks or blocks in a frame of image.

Preferably, the search blocks may be data blocks of a predetermined size in one frame of image, and the data blocks are spaced apart from each other by a certain distance.

Preferably, said step (2) can adopt the method of curve fitting to predict the global relative motion vector of current frame, specifically:

(20), according to the global relative motion vector of several frames before the current frame, determine a polynomial of degree p:

P(x)＝w _p x ^p +w _p-1 x ^p-1 +...+w ₀

Wherein, x is the relative frame number, P(x) is the global relative motion vector of the image frame whose relative frame number is x, and p is the order number that can describe the global relative motion vector trajectory, determined by the global motion mode;

(21) Using the polynomial calculation to obtain the global relative motion vector of the current frame, which is the predicted global relative motion vector.

Preferably, in the step (3), the method for judging whether there is shaking according to the difference of the global relative motion vector of the current frame obtained in the step (1) and the step (2) may specifically include:

(30), calculating the variance of the global relative motion vector of several frames before the current frame;

(31), if: |The global relative motion vector of the current frame obtained by prediction—the global relative motion vector of the current frame obtained by motion estimation|>K*the variance of the global relative motion vectors of several frames before the current frame, then it means that the current There is jitter in the frame, otherwise it is considered that there is no jitter, where 1≤K≤2.

Preferably, in the step (5), on the total pixel image of the current frame, according to the motion vector for compensating the jitter and the preset effective pixel image size, the method of clipping to obtain the effective image after the anti-shake compensation of the current frame Specifically can include:

(50), take the center position of the total pixel image of the current frame as the center point, and obtain the compensation center point according to the motion vector of the compensation shake;

(51), taking the compensation center point as the center, according to the size of the effective pixel image, determine the clipping area on the total pixel image of the current frame;

(52) Outputting the image pixels in the clipping area as an effective image after anti-shake compensation.

Preferably, the processing of the global relative motion vector in each step can be carried out from the horizontal direction and the vertical direction respectively, so that the global relative motion vector of the current frame obtained finally is composed of the horizontal component and the vertical component of the global relative motion vector. component, the obtained motion vector for shake compensation is also composed of its horizontal component and vertical component, and then judging the shake and cropping to obtain an effective image are also carried out from the horizontal direction and the vertical direction respectively.

Preferably, in the step (12), the frequency can be judged according to the comparison formula: the relative motion vector value with the highest frequency<the variance of the relative motion vector of the search block in the image*threshold, where 1≤threshold≤2 Whether the highest relative motion vector value exceeds the product of the variance of the relative motion vectors of the search block in the image and the given threshold.

In the above technical solution, the present invention proposes a video electronic anti-shake method and device, which perform motion compensation to eliminate shaking by performing motion estimation on image sequences and analyzing the motion of the shot scene. Compared with the prior art, the present invention does not need an external motion sensor, and realizes motion estimation and shake image compensation completely through digital image processing methods, and has the characteristics of simple, effective, low-cost, etc., and can be applied to digital video cameras, cameras, etc. in the collection device.

Description of drawings

Accompanying drawing 1 is the schematic diagram of a kind of video electronic anti-shake method of the present invention;

Accompanying drawing 2 is a schematic diagram of images before and after anti-shake compensation of a video electronic anti-shake method of the present invention;

Accompanying drawing 3 is the flowchart of a kind of video electronic anti-shake method of the present invention;

Accompanying drawing 4 is the block diagram of the circuit principle of the hardware of a kind of video electronic anti-shake method concrete implementation of the present invention;

Accompanying drawing 5 is the schematic diagram of the position relation of the three frame images after cutting among the present invention with respect to total pixel image;

Accompanying drawing 6 is a schematic diagram of a preferred motion estimation method adopted in the present invention;

Accompanying drawing 7 is a schematic diagram of an actual anti-shake example of the present invention.

Detailed ways

The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

The electronic anti-shake method proposed in this embodiment does not need to add a motion compensation sensor to obtain the motion state, but introduces the motion estimation method of video coding technology, and calculates the global relative motion vector of each frame through the motion estimation method, thereby obtaining the global relative motion vector of the current frame. The estimated value of the motion vector MV _N , and predict the global relative motion vector MV N of the current frame through the global relative motion vectors MV ₁ , MV ₂ , . . . , MV _{N -1} of several previous frames. If the difference between MV _N and MV _N is greater than a given threshold T, it is considered that shaking has occurred and motion compensation is required; otherwise, the original image is kept without compensation. Its principle is shown in Figure 1.

In order to realize electronic anti-shake in this embodiment, the acquisition pixel values of CCDs of all video acquisition devices cannot be utilized, so the effective number of pixels is less than the total number of pixels of CCD, which is an inevitable situation for realizing electronic anti-shake. The ratio of the number of effective pixels to the total number of pixels is related to the shake amplitude supported by the anti-shake. Assuming that the maximum jitter supported by this embodiment is taken as an index by the relative movement range, and the effective pixel area of a given CCD is the center of the total pixel area of the CCD, then the compensation range of the maximum jitter cannot exceed the area of the total pixel area of the CCD, so this embodiment The maximum compensation motion vector value in the horizontal direction is preset in XMAX, and the maximum compensation motion vector value in the vertical direction is YMAX, then the maximum compensation range in the horizontal direction is [-XMAX, XMAX], and the maximum compensation range in the vertical direction is [-XMAX, XMAX]. The maximum compensation range is [-YMAX, YMAX]. If the total range of jitter is within this range, compensation can be performed to implement electronic anti-shake; if it exceeds this range, electronic anti-shake will not be performed. The relationship between the total pixels of the CCD, the effective pixels and the effective pixels after anti-shake compensation is shown in Figure 2.

Referring to Fig. 3, the specific flow of a video electronic anti-shake method in this embodiment is as follows:

Step 1, adopting the motion estimation method to calculate the global relative motion vector of the current frame and several frames of video images before the current frame;

Step 2, predicting the global relative motion vector of the current frame according to the global relative motion vectors of several frames before the current frame;

Step 3, calculating the difference between the global relative motion vector of the current frame obtained in step 1 and step 2, and judging whether there is jitter according to the difference, when there is jitter, using the difference as a motion vector for compensating jitter, When there is no jitter, make the motion vector for compensating jitter 0;

Step 4. Judging whether the motion vector for compensating jitter is within the maximum compensation range determined by the preset maximum compensation motion vector value, and if it exceeds this range, use the maximum compensation motion vector value as the compensation jitter motion vector;

Step 5. On the total pixel image of the current frame, according to the shake-compensated motion vector and the preset size of the effective pixel image, crop and obtain the effective image of the current frame after anti-shake compensation.

In the above steps, the order of Step 1 and Step 2 can be reversed. In order to make the estimation and prediction of the global relative motion vector of the current frame accurate but with a relatively small amount of calculation, the F frame images before the current frame are used in this embodiment.

In the above steps, the video image used for motion estimation may be a total pixel image, or an effective image with an effective pixel size after compensation for dithering. Valid images are used in this embodiment.

This embodiment also provides a video electronic anti-shake device, including a motion estimation calculation module, a global relative motion vector calculation module, a global relative motion vector storage module, a motion vector calculation module for compensating jitter, and a jitter compensation module. Clipping module, global relative motion vector correction module and reference frame storage module.

The principle of the method of this embodiment will be described in detail below in conjunction with the schematic block diagram of the hardware circuit implemented in this embodiment shown in FIG. 4 .

(1) Pre-cutting module

The module input is the total pixel image of the current frame collected by the video acquisition module (such as CCD) of the video acquisition and output device and the global relative motion vector of each frame before the selected current frame, and the output is the actual displayed image size (i.e. effective pixel size) of the image.

Assume that the total pixel image size collected by the CCD is M×N, and the image size of effective pixels is M′×N′. We use MV _X (i) and MV _Y (i) to denote the relative motion vectors of the i-th frame image relative to the i-1-th frame image. When i=1, the effective pixel image is located at the center of the total pixel image, representing the first frame image in the video sequence, and the initial values are MV _X (1)=0 and MV _Y (1)=0 respectively. We also need the absolute displacement of the i-th frame relative to the total pixel center position (that is, the relative motion vector of the first frame), represented by SMV _X (i) and SMV _Y (i) respectively. This value is obtained by:

${\mathrm{<span\; class="notranslate">\; SMV</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; MV</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span>$

${\mathrm{<span\; class="notranslate">\; SMV</span>}}_{\mathrm{<span\; class="notranslate">\; Y</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; MV</span>}}_{\mathrm{<span\; class="notranslate">\; Y</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span>$

Absolute displacement plays an important role in the anti-shake operation of each frame image. The absolute displacement cannot exceed the constraint relationship between the total pixel and the effective pixel image, which can be expressed as:

-XMAX≤SMV _X (i)≤XMAX

-YMAX≤SMV _Y (i)≤YMAX

When the i-th frame of M×N size image is input, the center position of the clipping area is determined on the total pixel image of the current frame according to the SMV _X (i) and SMV _Y (i) of i-1, and then according to the preset effective pixel Image size Crop the current i-th frame image to form an M'×N' image. For the first frame of data, since SMV _X (1)=0 and SMV _Y (1)=0, the actual clipped area is an image of M′×N′ at the center of the image. FIG. 5 is a schematic diagram showing the positional relationship of the cropped three frames of images relative to the total pixel image. Among them, image A is the size of the total pixel image collected by the CCD, and its center position is point o; image B is the first frame image after cropping, and its center position coincides with the center position of image A; image C is the first frame after cropping (i-1) frame image, its center position is o′, and its absolute (that is, the relative displacement relative to the first frame) displacement relative to image B is SMV _x (i-1) and SMV _y (i-1) ; Image D is the cropped i-th frame (i.e. current frame) image, its central position is o ", its absolute displacement relative to image B is SMV _x i and SMV _y i, and its relative motion vector relative to image C is MV _X (i) and MV _Y (i).

(2) Motion estimation calculation module

The input of the motion estimation calculation module is the cropped M′×N′ i-th frame image and i-1-th frame image as reference frame images. The output is the relative motion vector of the i-th frame image relative to the i-1th frame image.

The motion estimation method can use the motion estimation method based on the video compression method, and perform motion estimation with blocks and macroblocks as search blocks. At this time, the relative motion vector of each macroblock or block in the image can be obtained through motion estimation, which can be used as the input of the global relative motion vector calculation module.

This embodiment recommends that the motion estimation module in video coding standards such as MPEG and H.26X be directly applied to the motion estimation module of this embodiment. These methods can be easily implemented and applied directly, but the disadvantage is that the motion estimation method in the video encoder usually has a large amount of calculation.

This embodiment also proposes a simplified motion estimation method. This method does not need to perform motion estimation on all image data blocks, which can greatly reduce the amount of calculation and improve the practical application value.

As shown in Figure 6, in this motion estimation method, the search block does not use the macroblock and block mode described in the video encoder. The search block in this method can be customized, usually 8×8, 16× 16 data blocks, and the selection of the search block is not the whole image, but some data blocks at certain intervals. These data blocks are searched for the best matching locations within a given region. The relative position to the reference position is the motion vector. The motion matching method can still use the search algorithm and matching criteria commonly used in video encoders, which are well-known technologies and will not be described in detail here.

(3) Global relative motion vector calculation module

The input of this module is the relative motion vectors of multiple search blocks obtained by the motion estimation module, and the output is the global relative motion vector of the current frame obtained by the motion estimation method.

Assume that the result of the search by the motion estimation module is L relative motion vectors. The frequencies of the horizontal and vertical motion vector magnitudes are counted separately, assuming that the relative motion vector values with the highest frequency are mvx _F and mvy _F respectively. Statistical analysis is performed on the relative motion vectors, and the variances are calculated as D _x and D _y , respectively.

The principle of obtaining the global relative motion vector through the motion estimation method in this embodiment is: to judge whether the current global relative motion exceeds the product of the variance of the relative motion vector of the block in the image and a given threshold, if the variance is relatively large, it means The motion state of is relatively complex, but the global relatively consistent motion is not obvious, on the contrary, it shows that the global relatively consistent motion is obvious, and there is a global relative motion vector.

The specific implementation is:

If mvx _F < D _x T _x , it indicates that the global relative motion vector in the horizontal direction of the current image is not obvious, and there is a horizontal component MVX(i)=0 of the global relative motion vector; otherwise, it indicates that there is a horizontal component MVX(i) of the global relative motion vector i) = mvx _F .

If mvy _F <D _y ·T _y , it indicates that the global relative motion vector in the vertical direction of the current image is not obvious, and there is a vertical component of the global relative motion vector MVY(i)=0; otherwise, it indicates that there is a vertical component of the global relative motion vector MVY( i) = mvy _F .

The given thresholds T _x and T _y are determined according to specific applications, and here it is recommended that 1≤T _x and T _y ≤2.

(4) Motion vector calculation module for jitter compensation

The input to this module is the global relative motion vectors MVX(i) and MVY(i) of the previous F-1 frame, where 1≤i≤F-1. The output is the shake compensated image motion vectors CMVX(F) and CMVY(F) for the current frame.

A commonly used method uses curve fitting, assuming that the data of the previous F-1 frame can determine a polynomial of degree P

P(x)＝w _P x ^P +w _P-1 x ^P-1 +...+w ₀

Here x is the relative frame number, we can choose the first frame of F-1 frame as 1, the second frame as 2, the F-1th frame as F-1, and the current frame as F. The output P(x) is the global relative motion vector in the horizontal or vertical direction. Here, p is set according to the actual situation, and is the order that can describe the trajectory of the global motion vector. For a variety of normal global motion modes, such as constant-speed aerial photography, changing from speed 0 to constant-speed aerial photography, etc., can be described by a P-degree polynomial. For example, when the global motion mode is uniform speed aerial photography, p takes 1, and when the uniform acceleration, p takes 2. Usually, when calculating, the p value is selected according to the actual supported sports requirements. The higher the p, the more complex the described sports will be. Generally, p≤5 can meet the usual sports requirements.

By fitting the image of the previous F-1 frame, the global relative motion vector of the current frame in the horizontal direction and the vertical direction is respectively predicted, and the predicted global relative motion vector of the current frame can be obtained as MVX(F) and MVY(F) .

At the same time, the variances of the global relative motion vectors of the previous F frames are calculated as: DMVX(F) and DMVY(F), and for the predetermined coefficient K, the motion vector for compensating the jitter can be determined by the following formula:

If |MVX(F)-MVX(F)|>K·DMVX(F), it means that there may be jitter in the current horizontal direction, and compensation is required, then CMVX(F)=MVX(F)-MVX(F) , otherwise it is considered that there is no jitter and it belongs to the normal moving range, then CMVX(F)=0.

If |MVY(F)-MVY(F)|>K·DMVY(F), it means that there may be jitter in the current vertical direction, and compensation is required, then CMVY(F)=MVY(F)-MVY(F) , otherwise it is considered that there is no jitter and it belongs to the normal moving range, then CMVY(F)=0.

The coefficient K here is recommended to be 1≤K≤2.

As mentioned above, since this embodiment can only perform effective shake compensation within the range of shake compensation, the judgment formula

-XMAX≤CMVX(F)≤XMAX

-YMAX≤CMVY(F)≤YMAX

Whether it is true, if it is true, it indicates that the total range of jitter is within the maximum compensation range, and directly output the horizontal component CMVX(F) and vertical component CMVY(F) of the motion vector for compensating the jitter. If one of them is not true or both are not true, that is, the maximum compensation range is exceeded in the horizontal direction and/or vertical direction, then the maximum compensation motion vector value XMAX and/or YMAX is used as the level of the motion vector for the compensation shake component and/or vertical component output.

(5) Jitter compensation module

The input of this module is motion vector CMVX(F) and CMVY(F) for compensating jitter and the current frame size is M×N total pixel image, and the effective pixel area of M′×N′ size at the corresponding position is cut according to the motion vector, thus The output is the effective pixel area after anti-shake compensation. Specifically:

First, take the central position of the total pixel image of the current frame as the center point (such as the o point in Figure 5), and obtain the compensation center point (such as the o "point in Figure 5) according to the motion vector of the compensation shake;

Then, take the compensation center point (such as o " point in Fig. 5) as the center, and determine the clipping area on the total pixel image of the current frame according to the effective pixel image size M' * N';

Finally, the image pixels in the clipping area are output as the effective image after anti-shake compensation.

(6) Global relative motion vector correction module

Due to the anti-shake compensation, the relative motion position of the Fth frame relative to the F-1th frame has changed, so the global relative motion vector of the Fth frame needs to be corrected.

This module inputs motion vectors CMVX(F) and CMVY(F) for compensating jitter of the current frame and the global relative motion vectors MVX(F) and MXY(F) of the current frame obtained by motion estimation, thereby correcting according to the following formula,

MVX(F)＝MVX(F)-CMVX(F)

MVY(F)=MVY(F)-CMVY(F),

and output to the global relative motion vector storage module for storage

(7) Global relative motion vector storage module

Because the calculation of the anti-shake compensation motion vector requires the global relative motion vector of the previous F frame. Therefore, the global relative motion vector of the previous F frame is saved through the Global Vector Memory.

(8) Reference frame storage module

Since the method of motion estimation is adopted, the image of the previous frame is required as a reference frame, and the image of the previous frame is stored in the reference frame storage module for motion estimation of the current frame image.

An actual anti-shake example is shown in Figure 7. The frame sequence of the digital video camera is Frame 1, Frame2, Frame 3, Frame 4. As the jitter occurs, compensation is performed continuously, but the compensation area is limited within the range of the total pixel area.

The advantage of this embodiment is that the anti-shake method based on motion estimation does not need an additional motion sensor device, but simply predicts the motion vector from the perspective of digital image processing to achieve anti-shake. This embodiment eliminates the bouncing phenomenon of the image caused by anti-shake when the camera moves and shoots.

The anti-shake technology in this embodiment is based on dynamic image anti-shake, and can be applied to cameras, digital video cameras, and other equipment.

Claims (9)

1. the method for a video electronic flutter-proof comprises the steps:

(1) adopt method for estimating calculating present frame and the present frame overall relative motion vectors of some frame video images before, be specially:

(10), the method for employing estimation is calculated the relative motion vectors of search block in each two field picture;

(11), the frequency of relative motion vectors amplitude of described each search block of statistics, obtain the relative motion vectors value that wherein frequency is the highest, calculate the variance of described relative motion vectors;

(12), judge whether the highest relative motion vectors value of described frequency has surpassed the variance of the relative motion vectors of search block and the product of given threshold value in the image, if do not surpass, the overall relative motion vectors that then shows current frame image is not obvious, and the overall relative motion vectors of putting present frame is 0; Otherwise, with the overall relative motion vectors of the highest relative motion vectors value of described frequency as present frame;

(2) according to the overall relative motion vectors of the overall relative motion vectors prediction present frame of some frames before the present frame;

(3) difference of the overall relative motion vectors of the present frame that obtains in calculation procedure (1) and the step (2), and judge whether to exist shake according to this difference, and when having shake, the motion vector of shaking by way of compensation with described difference, when not having shake, the motion vector that makes compensate for jitter is 0;

(4) judge that the motion vector of described compensate for jitter is whether in the determined maximum compensation range of predefined maximum compensating motion vector value, if exceed this scope, then with the motion vector of described maximum compensating motion vector value as described compensate for jitter;

(5) on total pixel image of present frame, according to the motion vector and the predefined valid pixel image size of described compensate for jitter, the effective image after the anti-shake compensation of cutting acquisition present frame.

2. method according to claim 1, it is characterized in that: the some frame video images before the present frame in described step (1) and the step (2) are effective image of each frame; And the video image of present frame carried out pre-cut, the method for its pre-cut specifically: at first obtain the absolute displacement with respect to total pixel image center of the effective image of former frame; Obtain the center, clipping region according to this absolute displacement value then, again according to the image size of predefined valid pixel, determine the clipping region with the center, clipping region, total pixel image to present frame carries out cutting at last, is the current frame video image of described valid pixel image size thereby obtain size.

3. as method as described in the claim 2, it is characterized in that: described search block is all macro blocks or the piece in the two field picture.

4. as method as described in the claim 2, it is characterized in that: described search block is the data block of the pre-sizing in the two field picture, and described data block keeps at a certain distance away each other.

5. as method as described in the claim 2, it is characterized in that: described step (2) adopts the overall relative motion vectors of the method prediction present frame of curve fit, is specially:

(20), according to the overall relative motion vectors of preceding some frames of present frame, determine a multinomial of p time:

P(x)＝w _px ^p+w _p-1x ^p-1+...+w ₀

Wherein, x is relative frame number, and P (x) is that relative frame number is the overall relative motion vectors of the picture frame of x,

P is determined by the global motion pattern for describing the exponent number of overall relative motion vectors track;

(21), adopt described polynomial computation to obtain the overall relative motion vectors of present frame, be the overall relative motion vectors that forecasting institute gets.

6. as method as described in the claim 5, it is characterized in that: in the described step (3), judge whether to exist the method for shake specifically to comprise according to the difference of the overall relative motion vectors of the present frame that obtains in step (1) and the step (2):

(30), the variance of the overall relative motion vectors of preceding some frames of calculating present frame;

(31) if: | the overall relative motion vectors of the present frame that the overall relative motion vectors-estimation of the present frame that obtains of prediction obtains | the variance of the overall relative motion vectors of preceding some frames of＞K* present frame, represent that then there is shake in present frame, otherwise think not have shake, wherein 1≤K≤2.

7. as method as described in the claim 6, it is characterized in that: in the described step (5), on total pixel image of present frame, according to the motion vector and the predefined valid pixel image size of compensate for jitter, the method for the effective image after the anti-shake compensation of cutting acquisition present frame specifically comprises:

(50), be central point with the center of total pixel image of present frame, obtain the compensation central point according to the motion vector of compensate for jitter;

(51), be the center with described compensation central point, according to described valid pixel image size, on total pixel image of present frame, determine the clipping region;

(52), the image pixel in the described clipping region of output is the effective image after the anti-shake compensation.

8. as method as described in the claim 7, it is characterized in that: in the described step (12) specifically according to comparison expression: the variance * threshold value of the relative motion vectors of search block in the highest relative motion vectors value＜image of frequency, 1≤threshold value≤2 wherein judge whether the highest relative motion vectors value of described frequency has surpassed the variance of the relative motion vectors of search block and the product of given threshold value in the image.

9. as method as described in the claim 8, it is characterized in that: the processing to overall relative motion vectors in described each step is all carried out from horizontal direction and vertical direction respectively, thereby the overall relative motion vectors of the final present frame that obtains is horizontal component and the vertical component by overall relative motion vectors to be constituted, the motion vector of the jitter compensation that is obtained also is made of its horizontal component and vertical component, and then to obtain effective image also be to carry out from horizontal direction and vertical direction respectively to the judgement of shake and cutting.

Images