CN101877790B - Panoramic video coding-oriented quick global motion estimation method - Google Patents

Abstract

The invention discloses a fast global motion estimation method for panoramic video coding. The method is divided into two steps: the first step is to roughly divide the local motion area and the global motion area with a luminance residual threshold model based on the combination of statistical characteristics and the threshold value, and obtain an approximate set of global motion estimation areas; the second step , using a motion vector residual grading threshold technique to gradually refine the global motion pixel set in the iterative process of energy residual function minimization, and finally separate the complete global motion region, so as to achieve the purpose of fast motion estimation. The advantages of the present invention are: the accuracy and robustness of the global motion estimation are taken into account, and the motion estimation calculation cost is greatly saved under the premise of ensuring that the quality of the reconstructed image is not lowered, but also slightly improved. The global motion estimation of panoramic video sequence has the effect of low algorithm complexity and accurate motion parameter estimation.

Description

A Fast Global Motion Estimation Method for Panoramic Video Coding

technical field

The invention relates to an image video coding compression technology, in particular to a fast global motion estimation method for panoramic video coding.

Background technique

Panoramic video is widely used in sports programs, 3D movies, multi-party video conferencing and other applications. A panoramic image is a six-sided or eight-sided fisheye camera that rotates or zooms around a fixed axis, and takes pictures of different orientations of the scene around the camera in the same time domain, and uses some "stitching" techniques for these pictures Seamless splicing, and then mapped into a cylindrical or spherical map according to the cylindrical or spherical mapping algorithm. Therefore, the resolution of the panoramic image is generally higher, and the motion details are richer. Moreover, since the panoramic video generally supports virtual camera motion reproduction in order to allow users to have a real scene experience, the interlacing feature of global motion and local motion is more obvious. Furthermore, the cylindrical and spherical mapping algorithms will cause object motion deformation, making motion estimation more difficult. But at the same time, we can find from the analysis of two-dimensional plane coordinates mapping cylindrical coordinates and spherical coordinates that the motion deformation of the panoramic image object on the cylindrical or spherical surface is mainly manifested as the deformation of camera rotation, scaling, and staggered movement. Therefore, We can use the global motion model to perform global motion estimation compensation on the panoramic video.

However, previous studies have shown that global motion estimation suffers from a computationally intensive problem. The key to global motion estimation is to estimate the parameters of the global motion model. Theoretically, for a definite global motion parameter model, such as an affine six-parameter motion model, only the motion vectors of three background pixels are needed to obtain the values of six motion parameters. However, the general motion vector method can only obtain motion vectors with limited precision. If only a few pixels of motion vectors are used to estimate the global motion model parameters, the accuracy of the obtained results is very low, so more pixel motion vectors are needed to participate in the global motion model. Motion model parameter estimation to improve accuracy. In this way, parameter estimation becomes a problem of solving contradictory equations, which can be solved by the least squares method, such as the Gauss-Newton iterative method. The solution equations are calculated through several iterations until the parameter values of the global motion model converge to a relatively stable value. However, this iterative process is the most computationally complex part of global motion estimation, especially when the global motion model to be processed is relatively complex and the number of iterations is large.

Furthermore, most video scenes contain global motion and local motion. If the motion vectors of all pixels are involved in global motion estimation, the pixels with local motion and pixels with large motion vector estimation errors will affect the global motion. Motion estimation produces a lot of interference, so that motion model parameter estimation must be iterated many times before it can converge to a stable value. It not only reduces the precision of global motion estimation, but also greatly increases the computational complexity.

The pixels in the local motion area are called outer points, and the global motion pixels are called inner points. The focus of the fast global motion estimation method we want to study is to effectively segment the local motion area from the whole image, so that the global motion area contains as many interior points and as few exterior points as possible. The key and difficulty in distinguishing outliers and inliers lies in the determination of the threshold T. In some literatures, T is set to a fixed value (called the fixed threshold method), but it is often difficult to find a suitable fixed threshold T that is competent for the entire iterative process: if T is too small, the iterative process is likely to start. Converge in the local area that requires local motion, especially when the local motion area accounts for a large proportion of the entire scene; and if T is too large, at the end of the iteration, some global motion pixels will be misjudged as local motion points, This greatly affects the accuracy of parameter estimation. Some literature mentions that the threshold T is replaced by a percentage threshold r% (called a fixed percentage threshold method), that is, the set of outliers is a set of r% pixels with a large residual value. However, since the proportion of the global motion area to the entire scene in different videos is different, this method can only obtain a part of the global motion area, or the set of inliers will contain many outliers. Some scholars have proposed an improved method: let the threshold T be the mean value of the residuals of all pixels (called the residual mean threshold method), but the residual mean method does not necessarily describe the global motion area and the local area well. The difference in the motion area, the experimental results also show that in some cases, this method may make the interior point set converge to the local motion area.

The above-mentioned inner region determination methods are difficult to balance the accuracy and robustness of parameter estimation, and often need to select an ideal threshold in advance for different video scenes, which has poor adaptability.

Contents of the invention

The purpose of the present invention is to provide a global motion estimation method for fast segmentation of global motion areas and local motion areas for panoramic video coding.

The technical scheme that the present invention solves the problems of the technologies described above is:

A fast global motion estimation method for panoramic video coding, the method is divided into two steps: the first step, the implementation of brightness residual threshold method; it further includes the following steps:

1-1. Divide the entire image into non-overlapping blocks with a block size of 16×16. Use a block matching algorithm to perform motion estimation on each block to obtain the motion vector (Δx, Δy) of each block;

1-2, use the motion vector (Δx, Δy) obtained in step 1-1 to perform motion compensation on each block of the current frame, and obtain the predicted value I _k of each pixel in the current frame;

以及在参考帧中的对应块内所有像素的梯度均值S′，

比较S和S′残差绝对值的大小，也即|S-S′|计算方式为公式(1)，即 $G_B=\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}|S_{i,j}-S_{i,j}^{\′}|/m\×n,$ m×n是块大小；1-3, calculate the gradient mean S of all pixels in the current block, where,

And the gradient mean S′ of all pixels in the corresponding block in the reference frame,

Comparing the absolute value of S and S′ residuals, that is, the calculation method of |SS′| is formula (1), namely $G_B=\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\underset{j=1}{\overset{\mathrm{no}}{\mathrm{\Σ}}}|S_{i,j}-S_{i,j}^{\′}|/m\×\mathrm{no},$ m×n is the block size;

式中，N是图像被划分成互不重叠的块的个数，G_i是根据公式(1)计算的图像中第i个块的亮度残差，代表当前帧的亮度残差均值；如果当前块的亮度残差G_B小于C_g，则当前块被判定为局部运动区域块，从全局运动宏块集合中剔除出去；1-4, calculate the threshold C _g , the threshold C _g is determined according to the formula (2), that is

In the formula, N is the number of blocks that are divided into non-overlapping blocks in the image, G _i is the brightness residual of the i-th block in the image calculated according to formula (1), Represents the mean value of the luminance residual of the current frame; if the luminance residual G _B of the current block is smaller than C _g , the current block is judged as a local motion area block and is removed from the global motion macroblock set;

The second step is to implement the motion vector residual classification threshold method, which specifically includes:

2-1, use the block matching algorithm to estimate the local motion displacement of each block in the current frame, and obtain the local motion vector (Δx, Δy) of each block;

2-2. Estimate motion model parameters: select the motion vectors of the center block with a size of 16×16 in three 48×48 image areas in the upper left corner, upper right corner, and lower left corner of the image as the first Gauss-Newton iterative model parameter estimation process Initialize the input value, through Gauss-Newton iterative calculation, to obtain the initialization of several motion model parameter values;

2-3. Calculate the global motion vector (Δx′, Δy′) of each block in the image: in the image, the block motion in the background area belongs to the global motion, while the block motion in the foreground area belongs to the local motion. The obtained motion model parameter values are resubstituted into the Gauss-Newton iteration, and the global motion vector of each block is calculated according to the coordinate (x, y) value of the central pixel point of each block. For the local motion block, the motion vector is Pseudo-global motion vector;

2-4. Calculating the motion vector residual: For the same block, combine the motion vector (Δx, Δy) of the block obtained in step 2-1 with the motion vector (Δx′) of the block obtained in step 2-3 , Δy′) as the variance;

2-5. Set the percentage threshold to exclude local motion blocks: use the percentage threshold P that is graded and decreased to gradually refine the global motion area, and set a motion vector residual threshold T that is graded and decreased to judge whether this iteration can end. In the iterative process, set the initial value of P to 50, and set T to 0.04 at the same time, the calculation formula is $\sqrt{{(\mathrm{\Δx}-{\mathrm{\Δx}}^{\′})}^2+{(\mathrm{\Δy}-{\mathrm{\Δy}}^{\′})}^2}>T;$

Mark the blocks that meet the above formula conditions and do not exceed the P percentage of the sum of the number of image blocks as local motion blocks, and the rest are marked as global motion blocks, which will be used as the block set for the next iteration of global motion estimation; if this global motion estimation If the motion vector residuals of all blocks in the block set do not meet the above conditions, this iteration ends, and the final motion model parameters are generated, and at the same time, the local motion area can be segmented according to the local motion block set marked in each iteration; P The decreasing rule of T and T: Before entering a new round of iterative calculation, update P to a quarter of P, and update T to T=T-0.01; after the first Gauss-Newton iteration, a new A set of motion model parameter values, and use this set of motion model parameter values as the input value of the next Gauss-Newton iteration;

2-6. Steps 2-3 to 2-5 are repeated until the motion vector residuals of all blocks are less than T or the number of iterations is equal to 4, then the iterative process ends.

Description of drawings

FIG. 1 is a flow chart of a fast global motion estimation algorithm for panoramic video coding in the present invention.

Detailed ways

The present invention will be further described below in conjunction with specific embodiment:

The specific implementation of the present invention is carried out in two steps:

1. The first step is to perform the brightness residual threshold method.

When using the recursive least square method to estimate the global motion, the initialization input value is the motion vector obtained after the block matching motion estimation of each macroblock in the image. The motion vector of a macroblock in the local motion area does not belong to the set of global motion internal points, and is an external point that interferes with the estimation of global motion parameters, which will consume a lot of computing time. Therefore, if the motion estimation is performed using the block matching algorithm When it is possible to predict the properties of the macroblock motion estimation, find and eliminate the macroblocks that may interfere with the global motion estimation in advance, which will inevitably greatly reduce the calculation amount of the global motion estimation.

The analysis of the video image shows that the texture of the global motion area is relatively rich; while the local motion is generally located in the area where the gray scale of the texture is consistent or the gray scale is extremely smooth. This is because the motion information contained in the local motion area is relatively small, and relatively speaking, the luminance residual between the original frame and the corresponding block of the reference frame is also very small. Therefore, the present invention pre-analyzes each macroblock through the luminance information after macroblock motion compensation to determine whether the macroblock belongs to a local motion macroblock, and if so, the macroblock is removed from the macroblock before global motion estimation. The global motion macroblock set is eliminated to save the calculation cost of the global motion estimation and improve the accuracy of the global motion estimation.

The algorithm is divided into the following four steps:

1) Divide the whole image into non-overlapping blocks, the block size is 16×16, use the block matching algorithm to perform motion estimation for each block, and obtain the motion vector (Δx, Δy) of each block. Theoretically, the smaller the block size, the more accurate the motion estimation result. In an extreme case, when the block size is one pixel, the motion estimation result is the most accurate. But in this way, the complexity of motion estimation will be greatly increased, and the number of motion vector bits that need to be coded and transmitted will also be more. This algorithm adopts 16×16 blocks, taking into account both the computational complexity and the accuracy of motion estimation. At the same time, for block matching motion estimation, we use three-step search fast motion estimation to further reduce the complexity of motion estimation. The motion vector (Δx, Δy) of each block obtained through the block matching algorithm needs to be saved as an initialization input for performing the second step motion vector residual classification threshold method.

的差异来判断当前帧像素(i，j)是全局运动区域像素还是局部运动区域像素。2) Use the motion vector (Δx, Δy) obtained by the block matching algorithm in the previous step to perform motion compensation for each block in the current frame, and obtain the predicted value I _k of each pixel in the current frame. Because (Δx, Δy) represents the inter-frame local motion displacement MV ₁ caused by the motion of the object itself, after inter-frame motion compensation, the brightness I _k of the current frame pixel (i, j) can be thresholded with the lower The brightness of the pixel at the position ((i, j)+MV ₁ ) in a frame

To determine whether the current frame pixel (i, j) is a pixel in the global motion region or a pixel in the local motion region.

以及在参考帧中的对应块内所有像素的梯度均值S′，其中，比较S和S′残差绝对值的大小，也即|S-S′|。如下式所示：3) Calculate the gradient mean S of all pixels in the current block, where,

And the gradient mean value S' of all pixels in the corresponding block in the reference frame, where, Compare the size of the absolute value of the residuals of S and S', that is, |SS'|. As shown in the following formula:

${\mathrm{<span\; class="notranslate">\; G</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

In the formula, m×n is the block size.

4) Calculate the threshold C _g . The threshold C _g is determined according to the following formula:

${\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; g</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; \&Center\; Dot;</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; N</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; G</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

代表当前帧的亮度残差均值。如果当前块的亮度残差G_B小于C_g，则当前块被判定为局部运动区域块，从全局运动宏块集合中剔除出去。考虑到阈值临界点C_g的取值过大将会导致全局运动参数估计失败，我们确定公式2中k的取值为：当低码率情况下，如编码CIF、QCIF格式的视频序列，k＝1.1；当高码率情况下，如编码4CIF、高清、全景视频序列，k＝1.2，也即如下式所示：In the formula, N is the number of blocks that are divided into non-overlapping blocks in the image, G _i is the luminance residual of the i-th block in the image (calculated according to formula 1),

Represents the mean value of the brightness residual of the current frame. If the luminance residual G _B of the current block is smaller than C _g , the current block is determined to be a local motion area block, and is excluded from the global motion macroblock set. Considering that the value of the threshold critical point C _g is too large will lead to the failure of the global motion parameter estimation, we determine the value of k in formula 2: when the code rate is low, such as coding video sequences in CIF and QCIF formats, k = 1.1; when the bit rate is high, such as encoding 4CIF, high-definition, and panoramic video sequences, k=1.2, which is shown in the following formula:

C _g ＝1.1G _av , in case of low bit rate

(3)

C _g ＝1.2G _av , in case of high bit rate

2. The second step is to implement the motion vector residual classification threshold method.

The principle of the motion vector residual method is mainly based on the difference between the global motion vector and the local motion vector, and the measurement of this difference is the key point of the method. The present invention adopts a classification threshold method to realize the accurate division of the measure. The main idea of this algorithm is: in the initial stage of iteration, first use a large percentage threshold to remove most of the outliers. Because at the beginning, the proportion of outliers that are quite different from inliers in the image is the largest (especially in local motion areas, which often have obvious differences in motion properties from global motion areas), a large The percentage threshold can quickly remove these outliers with large differences. Once the interference of outliers with large differences is excluded first, the entire iterative process converges faster; after the first iteration, a rough global motion has been obtained As a result of the estimation, the remaining outliers in the image at this time are pixels with small differences from the inliers, which account for a small proportion in the image. In the next few iterations, we can reduce the percentage threshold in stages value, gradually exclude these outliers with small differences, and refine the set of inliers precisely until as many outliers as possible are excluded from the set of inliers. The specific algorithm steps are as follows:

1) Estimate the local motion displacement of each block in the current frame using a block matching algorithm. This step has been performed in the previous link - threshold method based on luminance residual.

2) Estimate motion model parameters. Considering that the global motion area is generally concentrated in the image background area, that is, the edge part of the image, we select the three central blocks of the 48×48 image area (the block size is 16×16) in the upper left corner, upper right corner, and lower left corner of the image. The vector is used as the initialization input value of the first Gauss-Newton iterative estimation model parameter process, and several initial motion model parameter values can be obtained through Gauss-Newton iterative calculation.

3) Calculate the global motion vector (Δx', Δy') of each block in the image. In the image, the block motion in the background area belongs to the global motion, while the block motion in the foreground area belongs to the local motion, but we can still resubstitute the motion model parameter values obtained in the previous step into the Gauss-Newton iteration, according to each block center The value of the coordinate (x, y) of the pixel point is used to calculate the "global motion vector" of each block. For local motion blocks, this motion vector is a pseudo-global motion vector.

4) Calculate the motion vector residual. For the same block, take the motion vector (Δx, Δy) of the block obtained in step (1) and the motion vector (Δx′, Δy′) of the block obtained in step (3) as the variance, as shown in formula 4 shown. The purpose of calculating the motion vector residual is to judge the difference between the actual motion vector of the pixel block and the global motion vector. If the motion vector residual is large, it means that the block is likely to be a local motion block. Otherwise, the block is likely to be is the global motion block.

5) Set a percentage threshold to exclude local motion blocks. We use a progressively decreasing percentage threshold P to gradually refine the global motion area, and set a hierarchically decreasing motion vector residual threshold T to judge whether this iteration can end. During the first iteration, we set the initial value of P to 50. At the same time set T to 0.04, namely:

$\sqrt{{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&Delta;x</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&Delta;x</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&Delta;y</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&Delta;y</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; ></span>\mathrm{<span\; class="notranslate">\; T</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

Mark the blocks that meet the above formula conditions and do not exceed the P percentage of the sum of the number of image blocks as local motion blocks, and the rest are marked as global motion blocks, which are used as the set of blocks for the next iteration of global motion estimation. If the motion vector residuals of all blocks in the global motion estimation block set do not meet the above conditions, this iteration ends, and the final motion model parameters are generated, and at the same time, it can be divided according to the local motion block set marked in each iteration out of the local area of motion. The decreasing rule of P and T: Before entering a new round of iterative calculation, update P to a quarter of P (P=P/4), that is, during the second iteration, P is set to 12.5, and the first During the three iterations, P is set to 3.125... This is because as the number of iterations increases, the set of inliers tends to be more and more stable, and there are fewer and fewer outliers, so the P value needs to decrease linearly to avoid The iteration converges to a local set of interior points. At the same time, T is updated to T=T-0.01, because after the end of each iteration, the motion vector residuals of the remaining blocks are smaller, some of which are local motion blocks with smaller differences from the global motion attributes, and a Smaller T to tell them apart. When the iteration result tends to be stable, using decreasing T can simultaneously achieve the purpose of speeding up iteration convergence, reducing computational complexity and ensuring that the global motion estimation result is gradually refined. After the first Gauss-Newton iteration, a new set of motion model parameter values will be obtained. Use this set of motion model parameter values as the input values for the next Gauss-Newton iteration.

6) Steps 3), 4) and 5) are repeated until the motion vector residuals of all blocks are less than T or the number of iterations is equal to 4, then the iterative process ends. Experiments have proved that when the number of iterations exceeds 4, the accuracy of the solution of the model parameters increases extremely limited, that is to say, the results of the 4th, 5th, and 6th iterations are very similar, or even the same, and the solution of the model parameters has reached A limit of the accuracy of the classification threshold iterative method is set.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than limit them. Although the present invention has been described in detail with reference to the embodiments, those skilled in the art should understand that modifications or equivalent replacements to the technical solutions of the present invention do not depart from the spirit and scope of the technical solutions of the present invention, and all of them should be included in the scope of the present invention. within the claims.

Claims (1)

1. quick global motion estimation method towards panorama video code is characterized in that: this method was divided into for two steps and carries out:

1 first step is carried out the brightness residual threshold method; It further may further comprise the steps:

1-1 is divided into the piece of non-overlapping copies to entire image, and block size is 16 * 16, to the estimation of taking exercises of each piece, obtains the motion vector (Δ x, Δ y) of each piece with block matching algorithm;

1-2 uses the motion vector (Δ x, Δ y) that obtains among the step 1-1 to the compensation of taking exercises of each piece of present frame, obtains the predicted value I of each pixel of present frame _k

1-3, the gradient mean S of all pixels in the calculating current block, wherein,

And the gradient mean S ' of all pixels in the corresponding blocks in reference frame,

Compare the size of S and S ' residual absolute value, also be | S-S ' |, account form is formula (1), promptly $G_B=\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}|S_{i,j}-S_{i,j}^{\&prime;}|/m\&times;n,$ M * n is a block size;

1-4, calculated threshold C _g, threshold value C _gConfirm according to formula (2), promptly

In the formula, N is the number that image is divided into the piece of non-overlapping copies, G _iBe brightness residual according to i piece in the image of formula (1) calculating,

Represent the brightness residual average of present frame; If the brightness residual G of current block _BLess than C _g, then current block is judged as the local motion region unit, from the global motion set of macroblocks, rejects away;

Second step, carry out motion vector residual error classification thresholds method, specifically comprise:

2-1 uses block matching algorithm to estimate the local motion displacement of each piece of present frame, obtains the local motion vector (Δ x, Δ y) of each piece;

2-2; Estimate motion model parameters: the size of choosing the image upper left corner, the upper right corner, three 48 * 48 image-regions in the lower left corner is the initialization input value of the motion vector of 16 * 16 central block as Gauss's Newton iteration estimation model parametric procedure first time; Through Gauss's Newton iterative calculation, draw initialized several motion model parameters values;

2-3, the global motion vector of each piece in the computed image (Δ x ', Δ y '): in the image; The motion of the piece of background area belongs to global motion, and the motion of the piece of foreground area belongs to local motion, through in the motion model parameters value substitution Gauss again Newton iteration that obtains in the last step; (x, value y) calculate the global motion vector of each piece respectively based on the coordinate of each piece central pixel point; For the local motion piece, this motion vector is pseudo-global motion vector;

2-4, the calculating kinematical vector residual error: for same, the motion vector (Δ x ', Δ y ') of this piece that draws among the motion vector (Δ x, Δ y) of this piece that draws step 2-1 and the step 2-3 is made variance;

2-5; Percentage threshold is set gets rid of the local motion piece: adopt the percentage threshold P of hierarchical subtractive clustering to come progressively meticulous global motion zone; The motion vector threshold residual value T that a hierarchical subtractive clustering is set simultaneously judges whether this iteration can finish, and in first time iterative process, the initial value that P is set is 50; It is 0.04 that T is set simultaneously, and computing formula does $\sqrt{{(\mathrm{\&Delta;\; x}-{\mathrm{\&Delta;\; x}}^{\&prime;})}^2+{(\mathrm{\&Delta;\; y}-{\mathrm{\&Delta;\; y}}^{\&prime;})}^2}>T;$

Be labeled as the local motion piece to the piece of the P percentage that is no more than image block number summation at most that satisfies the following formula condition, the remaining global motion piece that is labeled as is as the set of blocks of overall motion estimation iterative computation next time; If the motion vector residual error of all pieces does not satisfy the following formula condition in this overall motion estimation set of blocks; Then finish this iteration; Generate final motion model parameters, simultaneously can be partitioned into the local motion zone according to the local motion set of blocks of each iteration institute mark; The rule of successively decreasing of P and T: before getting into the iterative computation of a new round, P is updated to 1/4th of P at every turn, T is updated to T=T-0.01; After Gauss's Newton iteration finishes for the first time, with obtaining a new cover motion model parameters value, the input value of this cover motion model parameters value as next round Gauss Newton iteration;

2-6, repeating step 2-3 to 2-5 equals 4 less than T or iterations up to the motion vector residual error of all pieces, then the finishing iteration process.

Images