KR20020044848A - Apparatus and method for block matching by using dispersed accumulate array in video coder - Google Patents

Abstract

PURPOSE: A block matching device is provided to use a dispersed accumulate array while estimating a motion in motion picture compression, thereby reducing calculations of a block matching algorithm. CONSTITUTION: A present/previous block memory(10) stores a present block and a previous block. A slice candidate extractor(20) performs an SAD(Sum of Absolute Difference) by slices of each candidate by the previous block, and extracts a candidate for the present block. A motion vector decider(40) decides a motion vector by using the extracted candidate. The slice candidate extractor comprises as follows. A candidate slice SAD calculator(21) calculates portions of SADs of predetermined candidates, and calculates a final SAD of the candidates. A candidate comparison extractor(22) removes candidates having a lower probability of being motion vectors by using the calculated SAD, and extracts a candidate having the lowest MAD(Mean Absolute Difference) value for finally-remaining candidates.

Description

Apparatus and method for block matching by using dispersed accumulate array in video coder}

The present invention relates to a block matching apparatus and a method using a spreading accumulation array of a video encoder, and more particularly, to a spreading accumulation of a video encoder which is suitable to reduce the computational amount of a block matching algorithm by using a spreading accumulation array in motion estimation. A block matching device using an array and a method thereof.

In recent years, interest in video encoding has been raised, and many standards such as H.26x, MPEG (Moving Picture Export Group) -1, MPEG-2, and MPEG-4 have been enacted.

These standards are applied to various applications such as video telephony, digital broadcasting, and high-definition TV. Most of them adopt complex motion compensation / discrete cod coding as video coding. Motion compensation is very important because it effectively compresses large amounts of digital data by effectively reducing redundancy between successive images. In particular, motion compensation in MPEG is performed differently according to the encoding method (P-picture or B-picture) of each frame.

For example, when the current frame is a P-picture, unidirectional prediction is applied, and in B-picture, unidirectional and bidirectional prediction are applied. Therefore, the importance of motion compensation increases, and the encoder requires a motion compensation mode in units of macroblocks having a size of 16x16 pixels.

Therefore, in order to implement the hardware (H / W), a frame memory for storing the pixel signal is required, and the process of reading the signal by storing the signal in this memory and moving the position by the motion vector, and considering the motion compensation and mode decision Is needed.

In addition, in order to determine a motion compensation mode in a video encoder, forward, backward, and bidirectional predictions must be performed, and a mode providing a minimum error among them according to a coding method of a current frame must be determined. For this real-time hardware implementation, three prediction tasks must be performed simultaneously during the allocated macroblock time.

In the MPEG video encoder, the processing unit of the image data is basically a macroblock unit having a size of 16 * 16 pixels. At this time, the process related to motion compensation includes writing the locally decoded macroblock data into the memory for the next frame, reading the data from the frame memory for prediction of the current macroblock, and using the read data. Performing predictions and determining the mode having the least prediction error.

That is, when the number of clocks allocated to the macroblocks is M, reading and writing data of one macroblock to the frame memory must be performed during this period. A common way to do this (described in more detail later) is to read during the initial M / 2 clock and read or write for the remaining M / 2 clock.

However, in a B-screen that allows bidirectional prediction, each prediction error is obtained by reading data for forward / reverse prediction, and bidirectional prediction data, which is an average of two predictions, must be obtained. Therefore, in this case, if the clock allocated to the macroblock is divided into the first half and the second half, the processing for bidirectional prediction becomes complicated.

1 is a block diagram of a general video encoder 1.

As shown therein, a frame memory 2 for storing data for a current image and a previous image as a macroblock; A motion estimator (3) for estimating the motion of the input image using the previous image obtained from the frame memory (2); And a motion compensator 4 for obtaining a predicted value of a block to be currently processed from a previous image stored in the frame memory 2 by using the motion vector obtained from the motion estimator 3.

The general video encoder 1 configured as described above first uses the previous image stored in the frame memory 2 to perform prediction with the input image in the motion estimator 3. Thus, motion estimation of the current image is performed using the previous image stored in the frame memory 2.

Therefore, the block matching method among motion estimation algorithms of the general encoder in the video encoder is to finally find the motion vector using the matching scale with the previous frame having a similar area as the block in the current frame. Therefore, the motion estimator 3 estimates a motion vector by receiving block information in each frame.

The motion vector obtained at this time is input to the motion compensator 4, and the motion compensator 4 extracts the motion compensation prediction value of the block to be processed currently from the previous image stored in the frame memory 2 using the motion vector. .

The basic unit for encoding is a macroblock of 16 * 16 pixels based on the luminance signal, and motion estimation, motion compensation, and mode determination are performed in this unit.

On the other hand, motion compensation has been widely used in the video compression field for a long time due to the high temporal similarity between successive frames of the input image. Motion compensation is divided into two steps as a method of suppressing redundancy of information existing between two adjacent frames in time.

First, motion estimation is performed to generate motion information between two frames, and divided into error prediction coding for encoding differential information between a current frame and a motion predicted frame.

The most commonly used method of motion estimation is a block matching algorithm (BMA). Block matching is an algorithm in which each block (usually 16 pixels wide and 16 pixels wide) of the current frame finds the most similar block in the previous frame between two consecutive frames.

In order to reduce the searching time and the amount of computation, the search is limited to searching only the area within the maximum search area M (N) + 2w around the current frame. Then, using a matching criterion, the relative position of the most similar block among the blocks in the maximum search area is set as a motion vector.

2 is a detailed block diagram of a motion estimation unit using conventional block matching.

As shown therein, a current / previous block memory 5 for storing a current block and a previous block; A SAD / MAD candidate extracting unit 6 for extracting a candidate for the current block by performing a sum of absolute difference (SAD) / mean absolute difference (MAD) on the previous block; The motion vector determination unit 7 determines a motion vector by using the candidate extracted by the SAD / MAD candidate extraction unit 6.

Therefore, in the conventional block matching method, after obtaining SAD values for preset candidates, each MAD is obtained so that the candidate having the lowest value becomes a motion vector.

3 is a configuration diagram of a previous / current frame in a search area used in FIG. 2.

In order to evaluate the accuracy of motion estimation, a mean absolute difference (MAD) is generally used, which is expressed by Equation 1 below.

Where M and N are the number of horizontal and vertical pixels of the block, and I _t (i, j) is the brightness value of the coordinate (i, j) in the frame at time t. Thus, I _t (i, j) represents one point of the current frame, and I _t-1 (i, j) represents any point in the previous frame. Therefore, MAD (x, y) becomes the screen brightness difference at the point located (x, y) away from the current frame.

And the MAD in the block (k, l) at the position moved by the horizontal k, vertical l is the following equation (2).

I has a value of 1 to 16, and j also has a value of 1 to 16. K and l are the values shifted from the current position to the right by w and left by w. MAD _(k-1) (x, y) is a value obtained by dividing MAD (x, y) by 256, which is a product of horizontal and vertical pixels M and N.

In addition, Sum of Absolute Difference (SAD) is used to reduce the use of dividers for hardware implementation. SAD differs from MAD in that it does not divide the total pixel value. This SAD is as shown in Equation 3 below.

In other words, SAD _{(k, l)} (x, y) is an absolute value accumulated between the difference between the value in the previous frame and the value in the current frame, and MAD is not divided by the total pixel value of 256.

Since the motion vector (MV) becomes the coordinate value of the position where the MAD value is the lowest in the search region, the following equation (4) is obtained.

Where arg is an argument function, (x, y) is a small group that is variable, (k, l) is a fixed value, and min is a minimum value. Therefore, MV (k, l) means that the position (x, y) where MAD has the smallest value is assumed to be MV.

Among the conventional block matching algorithms (BMAs), the full search BMA has the best performance because it searches all positions of the search area, but the H.261 and MPEG-1 video coders have 60% of the total throughput. It takes from 70% to 70%, which causes problems in hardware / software implementation.

Addition and subtraction of 2 * M * N are needed to compute the MAD of any position of the previous frame. If the search area is limited to w to reduce the amount of computation, the position to be searched in the fixed block of the current frame and the previous frame search area is (2w + 1) ² . If the size of the block is 16 in the forward direction and the search area w is 7, theoretically, a total of 57600 additions and subtractions are required, which requires a large hardware / software processing structure.

In order to compensate for the shortcomings of the calculation amount, fast search block matching algorithms (partial search BMAs) which have a performance close to the global search while reducing the calculation amount by searching some regions without searching the whole area have been proposed.

Assuming that the block has a horizontal size M and a vertical size N, it is necessary to subtract M * N and add M * N-1 for the final MAD of a candidate. So, for a square block of size 16, 256 subtractions and 255 additions are required. Therefore, even in the fast search matching algorithm, which has much less computation than the global search matching algorithm, a fixedly allocated 2 * M * N operation is required to calculate one candidate.

In summary, each candidate processes the subtraction of MN and the addition of MN-1 to obtain a final MAD value, as shown in Equation 3, to obtain a motion vector. In the case of global search, the position of the candidate having the lowest value becomes the motion vector after calculating the candidate of (2w + 1) ² , and in the case of the partial search, the candidates of the predetermined structure are calculated and the range is gradually increased. As you narrow down the search, the candidate with the lowest value becomes the final motion vector.

4 is a diagram illustrating an example of searching for a step reduction method in fast block matching using FIG. 2, and FIG. 5 is a diagram illustrating an example of searching for a step increment method in fast block matching using FIG. 2.

Thus, FIG. 4 is a three step search method for finding the lowest MAD (= SAD) in the symmetrical eight directions while reducing the w / 2 half of the search distance from the middle position.

In addition, Figure 5 is a method of increasing the step while moving the search model to the lowest candidate of the nine candidates using a square structure of 3x3.

Therefore, the motion vector can find the final MAD by calculating the MAD for each of the predetermined candidates according to a search method and then comparing them with each other.

FIG. 6 is a graph illustrating SAD trends of all candidates in the search region shown in FIG. 2.

Thus, each candidate performed S * of M * N (x-axis), and the final SAD value of MV is 972 (y-axis). The x-axis is the final SAD of 225 (= (2w + 1) ² ) values located at 256 (= M * N). Each line is a cumulative value of (k, l) blocks. Lowest SAD is the smallest difference between the current block and the previous block, which is determined as a motion vector. The SAD becomes large due to an abrupt SAD change of the candidates. That is, the SAD increases when the screen brightness is different by comparing the previous block with the current block.

As a result of the experiment, it can be seen that the trend of each candidate is suddenly changed (Abrupt SAD change) and intersected with each other, thereby failing to secure linearity. In this environment, it is difficult to find the exact criteria for removing candidates until the final SAD is obtained.

In addition, when the linearity of the SAD trend is not secured, the increase or decrease of the difference between the pixel values at the same position between two frames occurs irregularly. There is a decrease in linearity.

The prior art has not solved the problem of sudden change and linearity of these candidate trends.

Accordingly, the present invention has been proposed to solve the conventional problems as described above, and an object of the present invention is to spread the cumulative accumulation of a video encoder that can reduce the computational amount of the block matching algorithm by using the diffusion accumulation array during motion estimation in video compression. A block matching device using an array and a method thereof are provided.

1 is a block diagram of a typical video encoder.

2 is a detailed block diagram of a motion estimation unit using conventional block matching;

3 is a configuration diagram of a previous / current frame in a search area used in FIG.

4 is a diagram illustrating an example of a step reduction method in fast block matching using FIG. 2;

FIG. 5 is a diagram illustrating a search example of a step increment method in fast block matching using FIG. 2.

FIG. 6 is a graph showing SAD trends of all candidates in the search region shown in FIG.

7 is a block diagram of a block matching apparatus using spread accumulation array of a video encoder according to the present invention;

FIG. 8 is a detailed block diagram of a candidate extracting unit for each slice in FIG. 7.

9 is a flowchart illustrating a block matching method using spread accumulation array of a video encoder according to the present invention;

10 is a table showing each threshold matrix of the Bayer dither used in the present invention,

11 is a view comparing two-dimensional spatial sampling of a sequential accumulation method according to the prior art and a diffusion accumulation method according to the present invention.

12 is a graph showing the SAD trend of all candidates in the search region to which the present invention is applied.

13 is a table showing an example of the distance between the motion vectors between the initial / final slice to which the present invention is applied,

14 is a view showing a difference in slice configuration according to the prior art and the present invention,

15 is a graph showing the SAD trend in each slice to which the present invention is applied,

16 is a table comparing the performance of each block matching scheme according to the prior art and the present invention,

17 is a table comparing the time required for each block matching method according to the prior art and the present invention.

Explanation of symbols on the main parts of the drawings

10: current / previous block memory 20: candidate extracting unit for each slice

21: SAD calculation unit for each candidate slice 22: Candidate comparison extractor

31: memory 32: location generator

33: SAD calculation unit 34: comparison unit

40: motion vector determiner

Hereinafter, a block matching apparatus using a diffusion accumulation array of a video encoder and a method according to the technical spirit of the method will be described with reference to the accompanying drawings.

7 is a block diagram of a block matching device using a spread accumulation array of a video encoder according to the present invention.

As shown therein, a current / previous block memory 10 for storing a current block and a previous block; A candidate extractor 20 for each slice extracting a candidate for the current block by performing SAD (Sum of Absolute Difference) for each slice of each candidate by the previous block; And a motion vector determiner 40 for determining a motion vector using the candidate extracted by the slice candidate extractor 20.

The candidate extractor 20 for each slice may include: a candidate slice-specific SAD calculator 21 for calculating a final SAD of candidates by calculating a portion of SADs of candidates previously designated; A candidate comparison extractor which removes candidates having low probability as a motion vector for each slice by using the SAD calculated by the candidate slice SAD calculator 21 and extracts candidates having the lowest MAD value for the remaining candidates. It comprises a 22.

FIG. 8 is a detailed block diagram of a candidate extractor for each slice in FIG. 7.

As shown therein, a memory 31 for storing the accumulation order; A position generator 32 for generating a position of a mask and a position of an accumulated pixel using the accumulation order stored in the memory 31; A SAD calculator (33) for calculating SAD using values corresponding to the position generated by the position generator (32); And a comparison unit 34 for comparing the SADs among the candidates obtained by the SAD calculation unit 33 and dropping candidates with low probability.

In the above, the memory 31 stores the accumulation order using the diffusion accumulation method.

In the diffusion accumulation method, a Bayer Dither matrix is used.

9 is a flowchart illustrating a block matching method using spread accumulation arrays of a video encoder according to the present invention.

As shown therein, a first step (ST11) (ST12) of storing candidate positions and initializing candidates and slices; A second step (ST13) (ST14) for calculating whether the SAD for each slice is calculated for the candidate after the first step and calculating the final candidate; A third step ST15 of incrementing a candidate and returning to the second step if the last candidate has not been calculated; A fourth step (ST16) (ST17) of dropping candidates having a high SAD and determining whether the last slice has been reached once the last candidate has been calculated; A fifth step (ST18) of incrementing a slice and returning to the second step if the last slice has not been reached; If the last slice is reached, a sixth step ST19 of extracting the lowest SAD among remaining candidates is performed.

In the first step, the candidate position is determined using a diffusion accumulation method.

In the diffusion accumulation method, a Bayer Dither matrix is used.

Referring to the accompanying drawings, the operation of the block matching device and the method using the diffusion cumulative array of the video encoder according to the present invention configured as described above will be described in detail as follows.

First, if we make the trend of the SAD accumulation close to linear, we can predict the final MAD. Predicting the final MAD means that the motion vector can also be known. Therefore, the present invention proposes a method of making the transition curve shown in FIG. 6 close to a straight line.

7 is a block diagram of a block matching device using a spread accumulation array of a video encoder according to the present invention.

Thus, the current / previous block memory 10 stores macroblocks of the current frame and the previous frame.

The candidate extractor 20 for each slice extracts candidates for the current block by performing slice-specific SADs of the candidates of the previous block.

The motion vector determiner 40 determines the motion vector using the candidate extracted by the candidate extractor 20 for each slice.

The candidate extractor 20 for each slice includes a SAD calculator 21 and a candidate comparison extractor 22 for each candidate slice.

The candidate slice-specific SAD calculator 21 calculates a part of the SADs of the candidates that are specified in advance and calculates the final SADs of the candidates.

The candidate comparison extractor 22 removes candidates with low likelihood as a motion vector for each slice by using the SAD calculated by the candidate slice-specific SAD calculator 21 and has the lowest MAD value for the remaining candidates. Extract

Accordingly, the final SAD of the candidates is calculated by calculating a portion of the SADs of the candidates previously designated, thereby eliminating candidates that are unlikely as motion vectors. The candidates are removed in the same manner while the SAD calculation is performed on the remaining candidates in the same manner. The position of the candidate having the lowest MAD value for the remaining candidates becomes a motion vector.

FIG. 8 is a detailed block diagram of a candidate extractor for each slice in FIG. 7.

The memory 31 stores a cumulative order of a diffusion accumulation method using a Bayer dither matrix.

The position generator 32 generates the position of the mask and the position of the accumulated pixel using the accumulation order stored in the memory 31.

The SAD calculator 33 calculates the SAD using values corresponding to the position generated by the position generator 32.

The comparison unit 34 compares the SADs among the candidates obtained by the SAD calculation unit 33 and eliminates candidates having low probability.

Thus, a block match is entered using the values in the search area near the same position in the block and the previous frame in the current frame. Two-dimensional memory that stores the cumulative order to linearly increase the SAD of each candidate for estimating the final SAD in estimating the final SAD in order to estimate the final SAD. (31) is used. The memory 31 is used to generate the position of the mask and the position of the cumulative pixel. The partial SAD is obtained using the values corresponding to the generated position. And by comparing SADs between the candidates, candidates with low probability are eliminated.

9 is a flowchart illustrating a block matching method using spread accumulation arrays of a video encoder according to the present invention.

So, first, the candidate positions are stored in a diffusion accumulation method using a Bayer dither matrix.

The candidate and the slice are initialized.

SAD per slice is calculated for the candidate to the last candidate, and candidates with high SAD are dropped. This is done to the last slice to extract the lowest SAD of the remaining candidates.

This operation of the present invention will be described in more detail.

When accumulating SAD, we propose a distributed accumulate method rather than a conventional sequential accumulate method from the upper left to the lower right.

The diffusion accumulation method basically uses the Bayer Dither Matrix, which is used in halftoning. The matrix is composed of squares, with adjacent values in the matrix at the furthest distance.

10 is a table showing each threshold matrix of the Bayer dither used in the present invention. The mask is a memory for storing the order. Wow Is a memory that stores values of 4 and 4 horizontally.

The MAD of the cumulative accumulation method is expressed by Equation 8 below. s is the position of the current slice and S is the total number of slices. M _mask and N _mask are the horizontal and vertical sizes of the Bayer dither matrix.

In Equation 5, the x-axis relative coordinate i _mask and the y-axis relative coordinate j _mask are defined as in Equations 6 and 7 below.

And S, the total number of slices, is expressed by Equation 8 below.

S = M _mask * N _mask

Therefore, s _cur is a pixel position in the mask, d _cur is a position of 4x4 masks, and the final MAD MAD _{(k, l)} (x, y) is defined as in Equation 8 below.

Therefore, in the present invention, the linearity is secured by accumulating the difference between the values corresponding to the positions determined by the mask in the search region, thereby providing a foundation for predicting the final MAD.

FIG. 11 is a view comparing two-dimensional spatial sampling of a progressive accumulation method according to the prior art and a diffusion accumulation method according to the present invention.

11 illustrates the difference between the pixel reference position of the conventional technique and the present invention when SAD is accumulated by comparing pixels of M * 3 between two frames.

The conventional sequential accumulation method accumulates the values by subtracting pixel values of the current frame and the previous frame sequentially from the upper left to the lower right. In addition, the diffusion accumulation method proposed in the present invention is a method of accumulating the difference between values corresponding to positions which are not adjacent to each other in a two-dimensional space.

Therefore, when the object moves to an arbitrary position in the current frame in the progressive accumulation method, a difference occurs between each pixel of the previous frame and the current frame. Then, an abrupt SAD change occurs as shown in FIG. 6.

On the other hand, according to the present invention, since the SAD changes according to the prior art are divided at regular intervals in the cumulative trend curve by performing uniform sampling in two-dimensional space, linearity can be secured.

12 is a graph showing the SAD trend of all candidates in the search region to which the present invention is applied. That is, the results of using the present invention for the same block as in FIG. 6 are shown in FIG. From this, it can be seen that the trend curves increased linearly without a sudden change from the initial zero to the final SAD.

In FIG. 12, the candidate having the lowest value at 255 (= MN) of the x-axis (the x-axis is an increase in i, j, that is, the number of pixels to be subtracted in FIG. 6) is represented by the motion vector (MV). ) At this time, the value of SAD is 962. In FIG. 12, the number of candidates is (2w + 1) ² , and the trend of increasing SAD of each candidate becomes linearity, and thus, candidates having low MAD are finally fewer at any value on the left side of the x-axis with less computation. We can see that the probability of having a value is high.

Therefore, while obtaining SAD using Equation 9 for each candidate, the candidate having a low value keeps track of an increase in SAD, and the candidates having a relatively high value are eliminated to reduce the calculation amount. Here, the high SAD value means that the difference between pixel values, which is the difference between the current frame and the previous frame block, is large, and the difference in screen brightness, which is the pixel value, is large.

A slice increments by 1 every time M pixels are accumulated. Thus the slice increases from 1 to N.

FIG. 13 is a table showing an example of a distance between motion vectors between an initial / final slice to which the present invention is applied, and an Euclidean distance between two motion vectors having a SAD _MIN value in an initial slice (Slice 1) and a final slice (Slice 16). It can be expressed as shown in Equation 10 below.

13 is a ratio for each distance for a total of 131,340 blocks. This shows that, on average, the position of the motion vector of the initial slice (Slice 1) is the same as the final slice (Slice 16) in 62% of the blocks. Of 12 When added, the candidate having the minimum value of X at the point _M (full search 1/16 of the MAD operation amount) means that the average of 62% probability that the minimum value in X _{M * N} points.

As a result of analyzing the correlation, it can be seen that the point higher than X _M (slice 2 or higher) is higher than the probability of FIG. 13.

Therefore, if the linearity is secured, unlike the conventional method of subtracting all pixels of the current frame and the previous frame, that is, the pixels of the MN, and adding the result value, it is possible to predict candidates having a low final MAD with a small calculation amount.

In summary, the conventional method finds the lowest SAD value by reducing the range from a wide area to a narrow area by comparing the final MAD value, and the method according to the present invention detects the difference between pixels in the two-dimensional space. By accumulating, linearity is maintained to predict the final MAD, eliminating candidates with low probability, reducing computation, and predicting the probability of many candidates early, so that accurate motion vectors can be extracted even in blocks with large movements.

14 is a view showing the difference between the slice configuration according to the prior art and the present invention, and will be described for the area to be compared SAD according to the prior art and the present invention.

Since the conventional method uses the most final SAD, each candidate is compared as a result of accumulating the difference of pixels as much as MN, and thus has the same meaning as selecting the lowest candidate among the candidates in the rightmost part of FIG. 6. In terms of structure, the uppermost region is shown in the related art of FIG.

On the other hand, the present invention looks at the trend of increasing SAD and removes candidates, which is shown in FIG.

In this case, a total of 16 slices are formed, and candidates in each slice of FIG. 14 are candidates that span a vertical line designated as slice 16 (S16) to slice 1 (S1) of FIG. 15.

15 is a graph showing the SAD trend in each slice to which the present invention is applied.

By using this method, we proposed a fast block matching algorithm, and compared with the previous algorithms, the calculations used to calculate SAD were reduced by 30% to 60% and the computation time was also reduced by 30%. In terms of image quality, it has a performance close to that of the global search block matching method.

As described above, the present invention reduces the amount of computation of the block matching algorithm by using the diffusion accumulation array in motion estimation in motion picture compression.

Although the preferred embodiment of the present invention has been described above, the present invention may use various changes, modifications, and equivalents. It is clear that the present invention can be applied in the same manner by appropriately modifying the above embodiments. Accordingly, the above description does not limit the scope of the invention as defined by the limitations of the following claims.

As described above, the block matching apparatus and the method using the spread accumulation array of the video encoder according to the present invention can reduce the amount of computation of the block matching algorithm by using the spread accumulation array during motion estimation in video compression. .

In addition, the present invention can predict which candidates are close to the final MAD with 1/16 or 1/8 of the calculation amount by maintaining the linearity of the SAD. As we increase SAD, we can find motion vectors.

The present invention is a result of applying the fast block matching algorithm used in H.261, H.263, MPEG-1, MPEG-2, MPEG-4, etc., 1/20 of the global search method, 30 than the conventional fast block matching method. With a small calculation amount of about% ~ 60%, the result of the performance close to the global search was obtained. Accordingly, the present invention has laid the foundation for having a performance close to that of the global search or partial search (or fast) block matching algorithm with a small amount of computation in a block matching portion having a large amount of computation in a video encoder.

In order to analyze the performance of the proposed method, we used four types of bit-streams, each with different characteristics, and many of the existing papers compared with the well-known fast search algorithm. Therefore, the present invention also compared the performance with the basic fast block search algorithm TSS, NTSS, BBGDS, FSS, 2DLOG.

16 is a table comparing the performance of each block matching scheme according to the prior art and the present invention. Thus, the average number of times of using the MAD and SAD equations in the proposed block matching algorithms (FS, TSS, NTSS, BBGDS, FSS, 2DLOG) and the proposed method are compared. The average MAD shows better results than the existing fast block search algorithms and shows a result closer to the global search method. For various input images, the number of SADs is reduced by 20% to 64% than the average of the conventional algorithm. It can be seen that.

17 is a table comparing the time required for each block matching method according to the prior art and the present invention. In FIG. 17, the time required for performing the actual algorithm is compared. After optimization under the same conditions under the same conditions, the average execution time of each block matching method is investigated and shown in FIG. 17. Compared to the global search method (FS), 5.6 times, compared to the conventional fast block matching algorithms (TSS, NTSS, BBGDS, FSS, 2DLOG), the present invention (proposed) has a 20% to 33% time reduction effect. Able to know.

Claims (8)

A current / previous block memory for storing a current block and a previous block; A candidate extractor for each slice extracting a candidate for the current block by performing SAD for each candidate of the previous block; And a motion vector determiner configured to determine a motion vector by using the candidate extracted by the candidate extractor for each slice. The method of claim 1, wherein the candidate extract for each slice, A candidate slice-specific SAD calculator for calculating a final SAD of candidates by calculating a portion of SADs of candidates previously designated; A candidate comparison extractor which removes candidates having low probability as a motion vector for each slice and extracts candidates having the lowest MAD value for the remaining candidates by using the SAD calculated by the candidate SAD calculator. A block matching device using spread accumulation array of a video encoder. The method of claim 1, wherein the candidate extract for each slice, A memory for storing the accumulation order; A position generator for generating positions of masks and positions of accumulated pixels using the accumulated order stored in the memory; A SAD calculator which calculates SAD using values corresponding to positions generated by the position generator; And a comparator which compares SADs among candidates obtained by the SAD calculator to drop candidates with low probability. The method of claim 3, wherein the memory, A block matching device using a spreading accumulation array of a video encoder, characterized by storing a cumulative order using a spreading accumulation method. The method of claim 4, wherein the diffusion accumulation method, A block matching device using spread accumulation array of a video encoder, characterized by using a Bayer dither matrix. A first step of storing candidate positions and initializing candidates and slices; A second step of determining whether or not the final candidate is calculated by calculating the SAD for each slice of the candidate after the first step; A third step of increasing the candidate and then returning to the second step if the last candidate has not been calculated; A fourth step of, once calculating the last candidate, dropping candidates having a high SAD and determining whether the last slice has been reached; A fifth step of incrementing a slice and then returning to the second step if the last slice has not been reached; And a sixth step of extracting the lowest SAD from the remaining candidates when the last slice is reached. The method of claim 6, wherein the first step, A block matching method using spread accumulation array of a video encoder, wherein the candidate position is determined using a spread accumulation method. The method of claim 7, wherein the diffusion accumulation method, A block matching method using a diffuse accumulation array of a video encoder, using a Bayer dither matrix.