CN101815219A - Mobile evaluation method for real-time embedded multimedia design - Google Patents

Abstract

A method for performing a mobility assessment is provided. The method comprises selecting a current block in a current picture; obtaining motion vectors and residual value data of a plurality of adjacent blocks around a current block; setting a preset critical value according to residual value data of a plurality of adjacent blocks; comparing the current block with an initial reference block in a reference picture to obtain an initial comparison result, and comparing a predetermined critical value with the initial comparison result; if the initial comparison result is larger than a preset critical value, determining a predicted motion vector of the current block according to the motion vectors of a plurality of adjacent blocks; and comparing the blocks in the search window range corresponding to the predicted motion vector to find a corresponding reference block matched with the current block.

Description

Mobile Evaluation Methodology for Real-time Embedded Multimedia Design

technical field

The present invention relates to a method for performing mobility assessment, in particular to a method for performing mobility assessment that can reduce memory capacity and bandwidth.

Background technique

As the application of multimedia technology becomes more and more popular, the requirement of video compression technology becomes more and more important. Many video compression technical standards have been proposed one after another, and the current mainstream specifications include MPEG-4 and H.264/AVC. The basic principles of these standards are mainly to remove redundant data in image data, so as to reduce image storage space or image transmission volume. Motion Estimation is a very important part of video coding, which uses the similarity between consecutive pictures to remove the temporal redundancy of data, so as to achieve the purpose of data compression.

Fig. 1 is a schematic diagram of a block comparison algorithm often used in mobile evaluation. First, a current frame 100 with a frame size of W×H is divided into a plurality of blocks with a block size of N×N. Next, a search window (search window) with a size of (N+SR _H -1)×(N+SR _V -1) is set in the reference frame 110 (for example, the previous frame or the next frame). ) 112, and find a block 114 most similar to a current block (current block) 104 in the current frame 100 in the search window 112. Next, calculate the difference between the two blocks 104 and 114 and the motion vector 120 , and remove duplicate data by passing only the difference and the motion vector 120 . This step is motion evaluation. In other words, the purpose of motion estimation is to find the motion vector and error of each block in the current frame to represent the current frame. However, since motion estimation needs to compare many candidate blocks, this high computation will result in a large increase in memory bandwidth.

FIG. 2 shows the hardware architecture of the video encoding system 200, wherein the reference frame and the current frame are stored in the external memory 220, and the data required for motion evaluation are loaded into the internal memory 212 through the external bus 230 for the calculation engine (such as an embedded processor) 214 used. Therefore, when motion evaluation is performed, in order to perform data comparison operations, the required candidate block data in the search window of the reference frame will be transferred between the external memory 220 and the internal memory 212 through the external bus 230, thereby greatly increasing memory frequency. Width. Generally speaking, the size of the search window 112 is determined according to standards such as image resolution and/or compression specifications. The larger the search window 112 is, the more data needs to be loaded into the internal memory, and the required memory bandwidth is also larger. Therefore, it is necessary to provide a mobile evaluation execution method that can solve the excessive requirement of memory bandwidth.

Contents of the invention

In view of the existing problems in the prior art, the present invention provides a low-power and high-efficiency video coding method suitable for MPEG-4 and H.264/AVC for the square search (square search) algorithm, which can greatly reduce memory capacity and bandwidth, thereby reducing hardware costs, reducing power usage, and increasing execution speed.

According to one aspect of the present invention, a method for performing motion assessment is provided, comprising the following steps: selecting a current block in the current frame; obtaining motion vectors and residues of a plurality of neighboring blocks located around the current block Value data; set a predetermined critical value according to the residual value data of a plurality of adjacent blocks; compare the current block with the initial reference block in the reference frame to obtain an initial comparison result, and compare the predetermined critical value with the initial comparison Result; if the initial comparison result is greater than the predetermined critical value, the predicted motion vector of the current block is determined according to the motion vectors of multiple adjacent blocks; and the block comparison is performed in the search window corresponding to the predicted motion vector to find The corresponding reference block that matches the current block.

According to another aspect of the present invention, a computer-readable medium is provided for storing program instructions, wherein when the program instructions are executed on a computing device, the computing device will execute the above method.

Other aspects of the present invention will partly be stated in the subsequent description, and partly can be easily understood from the description, or can be known from the embodiments of the present invention. Aspects of the invention will be understood and achieved by use of the elements and combinations particularly pointed out in the appended claims. It should be understood that the foregoing summary of the invention and the following detailed description are for illustrative purposes only, and are not intended to limit the present invention.

Description of drawings

The drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and, together with the description, serve to explain the principle of the invention. The embodiments described herein are preferred embodiments of the invention, however, it must be understood that the invention is not limited to the arrangements and elements shown, wherein:

FIG. 1 is a schematic diagram of mobile evaluation using block comparison algorithm;

Fig. 2 shows the hardware architecture of a video encoding system;

FIG. 3 is a schematic diagram of performing motion assessment with a square search algorithm according to an embodiment of the present invention;

Figure 4 is a schematic diagram describing the use of square search for mobile assessment;

Figure 5 shows an example using raster scanning;

FIG. 6 shows four data reuse architectures for levels A to D of reference pictures; and

FIG. 7 shows a flowchart of a method for performing mobility assessment according to an embodiment of the present invention.

[Description of main component labels]

100 Current Screen

104 Current block

110 Reference screen

112 Search window

114 block

212 Internal memory

214 Computing Engine

220 External memory

230 bus

300 Current screen

302 Current block

310 Reference screen

312, 314 Blocks

320 Search Window

402, 404, 412, 414, 416 blocks

422, 424, 426, 428, 430 blocks

405 Square Graphics

500 screens

510, 511, 512, 513, 514 blocks

610, 620 Search window

612, 614 Blocks

622, 624 Block row

630, 640 Reference screen

632, 634 Search window

642, 644 Search window row

Detailed ways

The present invention proposes a dynamic evaluation method that can effectively reduce memory bandwidth and reduce internal memory (on-chip memory) requirements for the square search algorithm and cooperate with the data reuse architecture. Dependency, dynamically resizes the search window, replacing known dynamic evaluation methods that require loading the entire search window. In order to make the description of the present invention more detailed and complete, reference may be made to the following description together with the diagrams of FIGS. 3 to 7 . However, the devices, components and method steps described in the following embodiments are only used to illustrate the present invention, and are not intended to limit the scope of the present invention.

FIG. 3 is a schematic diagram of performing motion assessment using a square search algorithm according to an embodiment of the present invention. For a current block 302 with a size of N×N in the current frame 300, a surrounding frame corresponding to the position of the current block 302 in the reference frame 310 is shown. A search window 320 is displayed to find out the block most similar to the current block 302 in the search window 320 . The comparison method adopted in this embodiment is to calculate the SAD value of each candidate block in the current block 302 and the search window 320, and its calculation method is as follows:

$\mathrm{<span\; class="notranslate">\; SAD</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}^{\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}^{\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span>$

Cij represents the current block, and Rij represents a candidate block. In other words, the intensity of each pixel in the current block is subtracted from the intensity of each pixel in a candidate block, and then the absolute values of the resulting N×N differences are added to obtain the SAD value. The smaller the SAD value, the more similar the two blocks are. But it is worth noting that in this embodiment, although the SAD value is used as the judgment of the similarity with the current block 302, the method is not limited to this, and other comparison methods, such as mean square error (mean square error) or average Absolute error (mean absolute error) and the like are also applicable to the present invention.

The embodiment shown in FIG. 3 uses a square search method to find the block most similar to the current block 302 from the reference frame 310, wherein the square search algorithm uses a square search algorithm within the search window during the motion evaluation process. Graphics for comparison. FIG. 4 is a schematic diagram illustrating motion estimation using square search. First, in step 1, the block 402 corresponding to the current block position in the reference picture is taken as the starting point, and the most matching area is found in the square figure 405 composed of 9 blocks centered on the block 402 block (that is, the block with the smallest SAD value), in this example, it is assumed that the most matching block is block 404 . Then, in step 2, the block 404 is used as the center point, and the square pattern is used to continue searching for the most matching block. At this time, the data of the newly added blocks 412, 414, and 416 need to be loaded. Assuming that the most matching block in step 2 is block 416, then in step 3, centering on block 416, load the newly added blocks 422, 424, 426, 428, and 430 and repeat the above comparison operation . Assuming that the most matching block in step 3 is block 428, then in step 4, the above operations are repeated centering on block 428. If the most matching block in step 4 is at the center of the square figure, that is, if the most matching block is block 428, then the square search for this current block can be stopped.

Returning to FIG. 3 , in this embodiment, the comparison is first started from the block 312 corresponding to the position of the block 302, and the most matching area is found in the area composed of 9 blocks centered on the block 312 block (that is, the block with the smallest SAD). For example, nine blocks can be compared sequentially using the path shown by the arrows in FIG. 3 . Assuming that the block 314 has the smallest SAD value among the 9 blocks, the above steps are repeated with the block 314 as the center until the most matching block is at the center of the 9 blocks compared.

All the blocks in the current frame will go through the above-mentioned motion evaluation process to find the corresponding closest block in the reference frame, and the order in which the motion evaluation is performed will affect the motion evaluation of a specific block When , which surrounding blocks have already performed mobile evaluation. For example, FIG. 5 shows an example of using raster scanning. In this example, all the blocks in the frame 500 are scanned from left to right and from top to bottom. Therefore, when movement evaluation is to be performed on a certain block (such as block 510), the blocks on its left (511), upper left (512), upper (513), and upper right (514) have all been performed Through the procedure of motion evaluation, the motion vectors of these adjacent blocks and their corresponding SAD values are known. By combining the obtained adjacent block related data with the spatial correlation between the adjacent blocks, the movement vector of the current block can be predicted to dynamically adjust the search window range, and it can also be used to set the current block SAD threshold.

Taking the embodiment shown in FIG. 3 as an example, the predicted search window range and predetermined threshold of the current block 302 can be obtained according to the comparison results of adjacent blocks, as described below. First, obtain the motion vector and comparison data obtained after comparing adjacent blocks, wherein the comparison data is the SAD value (hereinafter referred to as residual data) of the matching block it searches for, and then set The predetermined threshold of the current block is as follows:

α _n =(2×LEFT _SAD +2×TOP _SAD +TOP-RIGHT _SAD +TOP-LEFT _SAD +ε)/6

Among them, LEFT _SAD represents the residual value data of the block on the left side of the current block, TOP _SAD represents the residual value data of the upper block of the current block, TOP-RIGHT _SAD represents the residual value data of the upper right block of the current block, and TOP-LEFT _SAD Represents the residual value data of the block in the upper left corner of the current block, and ε represents the constant factor of the current block. ε is a compensation coefficient that can be fine-tuned, and its selection is the application of empirical rules, and the designer can adjust it according to the actual application. In this embodiment, considering the distance between the blocks, the weights of the left and upper blocks are doubled. However, in other embodiments, the weights of adjacent blocks can be adjusted according to actual applications. It should be noted that the present invention is not limited to the raster scanning order shown in FIG. The motion evaluation results of neighboring blocks that can be obtained for a block are used to predict the range of the search window.

In the embodiment of FIG. 3 , when performing motion evaluation on the block 302 , the block 312 corresponding to the position of the block 302 in the reference frame 310 is first loaded into the internal memory, and then the block 302 and the block 312 are compared. And calculate its SAD value. If the SAD value is smaller than the predetermined critical value α _n , the motion evaluation of the block 302 can be ended (the motion vector is (0, 0)), and the motion evaluation of the next block can be performed. In this way, the memory bandwidth (Memory Bandwidth) can be reduced to only about 11% for static movies.

If the SAD value between the block 302 and the block 312 is greater than the predetermined threshold α _n , then load the predicted search window range for subsequent comparison. The prediction of the search window range can be determined by using the obtained adjacent block related data and the spatial correlation with the adjacent blocks. In this embodiment, the motion vector of the current block is predicted by using the motion vectors of the upper left block 1 , the upper block 2 , the upper right block 3 and the left block 4 of the current block. First, define the coordinates corresponding to the center point of the current block as (0, 0), and the corresponding coordinates of the center points of the upper left block 1, upper block 2, upper right block 3, and left block 4 of the current block are respectively is (-16, 16), (0, 16), (16, 16), (-16, 0). Next, according to the least squares method, use the movement vectors and relative coordinates of the upper left corner block 1, upper side block 2, upper right corner block 3 and left block 4 to obtain the most suitable regression plane (regression plane): z= c-ax-by, to predict the movement vector of the current block. Substituting the coordinates and moving vectors of the four adjacent blocks, we can get:

$\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 4</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; (</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; ax</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; by</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; MV</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}$

Take the partial differentials of a, b, and c respectively:

$<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; /</span><span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ;</span>$

$<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; /</span><span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ;</span>$

$<span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; /</span><span\; class="notranslate">\; \&PartialD;</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ;</span>$

From the above three formulas, the values of a, b, and c can be obtained as follows:

a=1/32(MV ₁ -MV ₃ );

b=1/96(-5MV ₁ -2MV ₂ +MV ₃ +6MV ₄ );

c=1/2(-MV ₁ +MV ₃ +2MV ₄ );

Substituting the values of a, b, and c and the current block coordinates (0, 0), the current block movement vector can be predicted as:

MV＝1/2(-MV ₁ +MV ₃ +2MV ₄ )

The range of the search window can be predicted according to the predicted movement vector of the current block and its corresponding square graphic area, so that only part of the required data can be downloaded, avoiding downloading the data of the known entire search window. After the dynamically adjusted search window is loaded, the comparison is performed in the above-mentioned square search method within the search window. To sum up, the present invention only loads the data of the reference picture corresponding to the original position of the processed block at the beginning, and then dynamically adjusts how much data needs to be loaded into the internal memory after comparison, that is, the size of the search window can be dynamically determined . Therefore, the present invention can reduce the amount of data to be loaded into the internal memory, not only reduce the time and power consumption of data transmission, but also reduce the size of the required internal memory and reduce hardware costs.

In addition to using the method of data prediction, the present invention also applies a data reuse framework in memory management, and reduces the times of memory access and data transfer by temporarily storing reused data in the internal memory. In other words, after analyzing the reusability of data, some data is avoided to be repeatedly accessed by adding an internal memory, thereby reducing memory bandwidth requirements. For the relevant description of the data reuse architecture, please refer to "Having Minimum Memory Bandwidth" published by D.X.Li et al. in IEEE Trans.Consumer Electron., vol.53, no.3, pp.1053-1060, Aug.2007 Architecture Design for H.264/AVC Integer Motion Estimation with Minimum Memory Bandwidth (Architecture Design for H.264/AVC Integer Motion Estimation with Minimum Memory Bandwidth)", by J.C.Tuan et al. in IEEE Trans.Circuits Syst.Video Technol., vol.12, No.1, pp.61-72, "On the data reuse and memory bandwidth analysis for full-search block-matching VLSI architecture" published in Jan.2002 architecture)", published by C.Y.Chen et al. in IEEETrans.Circuits Syst.Video Technol., vol.16, no.4, pp.553-558, Apr.2006 "for mobile evaluation with corresponding coding order Level C+ data reuse scheme (LevelC+data reuse scheme for motion estimation with corresponding coding orders)", and, by T.C.Chen et al. in IEEE Trans.Circuits Syst.Video Technol., vol.17, no.2, pp .242-247, "Single Reference Frame Multiple Current Macroblocks Scheme for Multiple Reference Frame Motion Estimation in H.264/AVC published in Feb.2007 "Single Reference Frame Multiple Current Macroblocks Scheme for Multiple Reference Frame Motion Estimation inH. 264/AVC)", the contents of which are incorporated herein by reference.

The performance of the data reuse architecture can be evaluated by the following two factors: the size of the internal memory and the redundant access parameter Ra, where the internal memory can be used to represent the memory size required to temporarily store reference data for data reuse, and the redundant memory The parameter Ra can be used to evaluate the bandwidth of the external memory, which is defined as follows:

The lower the degree of data reuse, the larger the Ra value, and requires more memory bandwidth. Conversely, the higher the degree of data reuse, the smaller the Ra value, and requires less memory bandwidth. The total memory bandwidth BW can be expressed as follows:

BW＝f×W×H×(Ra _{current picture} )+f×W×H×(Ra _{reference picture} )

Where f is the picture update rate, W is the picture width, and H is the picture width.

Generally speaking, memory bandwidth depends on the frame rate, frame size, search window size, and Ra value, etc., and for specific video compression applications, the frame rate and frame size are usually fixed values, so The present invention reduces the size of the search window by selecting a data reuse architecture with a smaller Ra value and using a data prediction method, thereby effectively reducing memory bandwidth.

For the current picture, each block will be accessed SR _H x SR _V times on average, namely

But as long as the internal memory with the size of NxN is added, the Ra of the current picture can be reduced to 1, as follows:

As for the reference picture, FIG. 6 shows four data reuse structures of levels A to D for the reference picture, wherein the hatched part is the data that can be reused. Levels A and B are data reuse within a single search window 610, 620, respectively, and levels C and D are data reuse within different search windows. Specifically, for a block of size NxN pixels in the current frame, class A reuses a single search window of size (N+SR _H -1)×(N+SR _V -1) in the reference frame In 610, the overlapping pixels between two consecutive candidate blocks 612 and 614 in the horizontal direction, while the level B is to reuse the overlapping pixels between the two consecutive rows of candidate blocks 622 and 624 in the vertical direction in the search window 620 pixels. Level C is to reuse the overlapping pixels between the search windows 632 and 634 corresponding to two horizontally continuous blocks in the reference frame 630 , and level D is to reuse the two rows of vertically continuous blocks in the reference frame 640 The overlapping pixels between the search windows 642 and 644 corresponding to the respective blocks. As mentioned above, the total memory bandwidth depends on Ra, and the Ra of the grade A to D architecture can be calculated as follows:

Grade A:

Grade B:

Grade C:

Grade D:

On the other hand, as can be seen from Figure 6, the internal memory size required for the grade A to D architecture is as follows:

  Reuse level Internal memory size A N×(N-1) B (N+SR _H -1)×(N-1) C (SR _H -1)×(SR _V +N-1) D W×(SR _V -1)

From the above, it can be seen that the smaller the size of the internal memory, the greater the memory bandwidth requirement (such as grade A). On the contrary, although the memory bandwidth requirement of the grade D architecture can be greatly reduced, relatively larger internal memory size is required .

Class C can achieve a better balance between internal memory size and external memory bandwidth, but with the improvement of video resolution, the data reuse architecture of Class C is no longer sufficient for practical applications. Therefore, the present invention combines the two functions of data reuse and data prediction. In the level-C framework, the size of the search window of the current block is predicted and dynamically adjusted according to the size of the search window of adjacent blocks, replacing the original level-C framework. It is necessary to load the overlapping areas of the entire search windows corresponding to two consecutive blocks in the horizontal direction, and only need to load the corresponding predicted search window range. In this way, not only the size requirement of the internal memory can be effectively reduced, but also the memory bandwidth required by the class-C architecture can be further reduced.

FIG. 7 shows a flowchart of a method for performing mobility assessment according to an embodiment of the present invention. Generally speaking, when executing the block comparison motion evaluation algorithm, the current frame is first divided into multiple blocks, and the scanning sequence for performing motion evaluation on the multiple blocks is determined. In this embodiment, each block is sequentially evaluated for motion in a raster scan order, and the motion evaluation algorithm for each block is a square search algorithm. First, in step S700, one of the blocks is selected for motion evaluation, and the dynamic scanning window range of its adjacent blocks, the motion vector prediction value of the adjacent blocks, and the comparison data of the adjacent blocks are obtained. Next, in step S710, the predetermined critical value α _n of the current block is defined according to the residual value data of adjacent blocks _. Add the resulting value. It should be noted that the calculation of the predetermined critical value α _n can be selected and adjusted appropriately according to actual application requirements. Next, in step S720, the data corresponding to the position of the current block in the reference frame is loaded, and a real comparison operation is performed with the current block to calculate its SAD value.

Next, in step S730, compare the predetermined threshold α _n set in step S710 with the SAD value obtained in step S720. If in step S730, the SAD value obtained by comparison is smaller than the predetermined critical value α _n , the procedure proceeds to step S740, ends the movement evaluation of this block, and then proceeds to step S750, to determine whether all blocks have been moved Evaluate. If it is judged in step S750 that all blocks have completed the motion evaluation, then the program proceeds to step S760 to end the motion evaluation of the current screen, if it is judged in step S750 that not all blocks have completed the motion evaluation, then the program returns to step S700, select Next A block continues to perform mobile evaluation.

In step S730, if the SAD value is greater than the predetermined threshold α _n , the process proceeds to step S770, where the motion vector of the current block is predicted by using the known motion vectors of adjacent blocks. In this embodiment, the prediction formula is obtained by calculating the regression plane by the least square method according to the moving vectors and coordinates of the upper left, upper, upper right, and left blocks. It should be noted that the prediction formula may be adjusted according to the scanning order. Next, in step S780, the square search area corresponding to the predicted current block motion vector is loaded from the external memory into the internal memory. Next, in step S790 , in the loaded search window, a square search algorithm is used to perform motion evaluation on the selected blocks to find the best matching block. After finding the best matching block in step S790, the process returns to step S750 to repeat the above steps until the motion evaluation of the current frame is completed. It should be noted that if a matching block cannot be found by square search in step S790, the block with the smallest SAD value can be selected as the matching block of the current block in the predicted search window, or other blocks can be loaded. Or a larger search window range to perform the comparison operation again, but the present invention does not limit the practicable manner in this case.

The present invention uses the residual data of adjacent blocks and the result of the first real comparison to reduce the required memory bandwidth and internal memory size, and predicts the current block using the motion vectors of the surrounding blocks The range of the search window. The present invention replaces the need to download the search ranges of two adjacent blocks in the horizontal direction instead of the level C architecture, and only needs to download the predicted search window range. Therefore, the present invention only needs to use 30% to 60% of the original internal memory, and the memory bandwidth only needs about 40% to 80%, which greatly reduces the internal memory and memory bandwidth. The mobile evaluation method combined with data prediction and data reuse framework proposed by the present invention can be applied to software and hardware memory management of various real-time embedded multimedia systems.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the scope of the claims of the present invention; all other equivalent changes or modifications that do not deviate from the spirit disclosed in the present invention should be included in the above-mentioned within the scope of the claims.

Claims (10)

1. one kind in order to carry out the method that moves assessment, comprises following steps:

A. in present picture, select present block;

B. obtain the motion-vector and the residual value data that are positioned at this present block a plurality of adjacent block on every side;

C. according to this residual value data setting predetermined critical of this a plurality of adjacent block;

D. compare this present block with initial reference block in reference picture and obtain initial comparison result, and relatively this predetermined critical and this initial comparison result;

E. if this initial comparison result greater than this predetermined critical, then determines the prediction motion-vector of this present block according to this motion-vector of these a plurality of adjacent block; And

F. in to the search window scope that should predict motion-vector, carry out the block comparison, to seek the corresponding reference block that is complementary with this present block.

2. method according to claim 1, wherein this initial reference block position in this reference picture is to the position of block in this present picture at present.

3. method according to claim 1, wherein step f is to use square search algorithm to carry out the block comparison.

4. method according to claim 1 is wherein in steps d, if this initial comparison result less than this predetermined critical, is this correspondence reference block with this initial reference block then.

5. method according to claim 4 also comprises at all block repeating step a to f in this present picture, and wherein step a selects this present block with grating scanning mode.

6. method according to claim 1, wherein this initial comparison result is the function of the absolute error sum total between this initial reference block and this present block.

7. method according to claim 1, wherein this predetermined critical is to be set as follows:

α _n＝(2×LEFT _SAD+2×TOP _SAD+TOP-RIGHT _SAD+TOP-LEFT _SAD+ε)/6；

LEFT wherein _SADBe the residual value data of this present block left side adjacent block, TOP _SADBe the residual value data of this present block top adjacent block, TOP-RIGHT _SADBe the residual value data of this present block upper right corner adjacent block, TOP-LEFT _SADBe the residual value data of this present block upper left corner adjacent block, ε is a constant coefficient.

8. method according to claim 1, wherein this prediction motion-vector of this present block is:

MV＝1/2(-MV ₁+MV ₃+2MV ₄)；

MV wherein ₁Motion-vector, MV for this present block upper left corner adjacent block ₃Motion-vector, MV for this present block upper right corner adjacent block ₄Motion-vector for this present block left side adjacent block.

9. method according to claim 1 also comprises following steps:

G. from an external memory storage this search window scope is loaded, and this search window scope is stored in the internal storage;

H. the horizontal direction that determines the present block that this is selected is to the pairing prediction search window of right continuous block scope; And

I. with step h compared to step g loaded newly-increased prediction search window scope, this external memory storage is loaded in this internal storage certainly.

10. a computer fetch medium in order to stored program instructions, wherein when this program command is executed on the calculation element, will make this calculation element enforcement of rights require 1 described method.

Images