KR100286818B1 - Fast motion estimating method for real-time video coding - Google Patents

Abstract

The present invention relates to a fast motion prediction method for real-time video compression in an apparatus for compressing video data. According to the present invention, a three-step search method using a single mode error surface assumption (UESA) can reduce the number of search points. In this improved method, the high-speed motion prediction is possible by applying the half-stop method with a threshold and partial error for matching error to the improved method. It is possible to apply to software based real-time system, because it is possible to make high-speed calculation by reducing search point while maintaining almost the same as predicted picture quality by three-stage retrieval method.

Description

Fast Motion Prediction Method for Real-time Image Compression {FAST MOTION ESTIMATING METHOD FOR REAL-TIME VIDEO CODING}

The present invention relates to a fast motion prediction method for real-time image compression in an apparatus for compressing image data. In particular, the three-stage retrieval method using a single mode error surface assumption (UESA) can be improved to reduce the number of search points. In addition, the present invention relates to a fast motion prediction method for real-time image compression, in which a half-stop method having a threshold and a partial error for a matching error is appropriately applied to the improved method.

In general, video data has a huge amount of data compared to audio and text data, and thus real-time processing is impossible unless this video data is compressed.

By compressing the image data in a predetermined manner as described above, real-time processing of the image signal is possible in storage and transmission. In the international standard for compressing the current image, JPEG is used for the still image standard, and television is used for the video standard. MPEG1, MPEG2 used for satellite broadcasting, and MPEG4 currently under development for low bit rate transmission.

On the other hand, the compression of the image data is achieved by eliminating duplicate data, which is due to the difference between the data corresponding to the image information and the data representing the image information.

Such data duplication includes spatial overlap, stochastic overlap, and temporal overlap between frame images. First, spatial overlap results from the similarity of values between adjacent pixels. The neighboring pixel values are similar to each other. This spatial overlap processing uses a discrete cosine transform (DCT), which transforms the information of the image into the upper left of the frame image.

Next, the probabilistic overlap is due to the similarity of the symbols, which means that the data are not evenly distributed evenly, and that any symbol has a similar value to neighboring thimbles, and the processing of the stochastic overlap is entropy. Variable length coding (Variable Length Coding) is used, and the variable length coding allocates bits having a size proportional to the size of the thimble.

Finally, temporal overlap is due to the similarity between the current frame image and the previous frame image, and the processing of temporal overlap uses motion estimation / motion compensation (ME). Detects the motion vector between the current frame image and the previous frame image, creates a new frame image from the motion compensation using the detected motion vector, and then subtracts the frame image generated from the current frame image to generate the current frame image and the generated frame. Remove the same data between the images.

FIG. 1 is an overall configuration diagram of a general video data transmission / reception system. Referring to FIG. 1, a video data transmission / reception system compresses and transmits video data, and a video signal transmitted from the video data transmitter 100. It consists of a satellite (1) for receiving and transmitting to the receiver, and a video data receiver (200) for receiving the video signal from the satellite (1), decompressing, and restoring the original video data.

The image data transmitter 100 includes an MPEG source encoder 110 for extruding image data and sound data, a text encoder 130 for compressing text, and channel information in the encoded data to solve the noise problem. And a channel encoder 150, and an RF unit 170 for modulating the encoded data.

In addition, the video data receiver 200 includes a baseband processor 210 for restoring the video signal from the satellite 1 to the video data of the baseband through carrier removal, and the video data of the baseband processor 210. The channel decoder 220 performs error detection, correction, and reconstruction, and the MPEG decoder 230 decompresses the image data of the channel decoder 220 and restores the original image data.

FIG. 2 is an internal block diagram of an MPEG source encoder included in the transmitter shown in FIG. 1.

Referring to FIG. 2, the MPEG source encoder includes an 8 * 8 blocking unit 111 for dividing data of an input frame image into 8 * 8 blocks, and the current frame image from the 8 * 8 blocking unit 111. A subtractor 112 for subtracting the generated frame image, an 8 * 8 discrete cosine transformer 113 for performing discrete cosine transform on the current frame image from the subtracter 112, and an 8 * 8 discrete cosine transformer ( An 8 * 8 quantizer 114 for quantizing the frame image from 113, a variable length coding unit 115 for performing variable length coding on the current frame image from the 8 * 8 quantizer 114, Inverse discrete cosine transform is performed on the 8 * 8 inverse quantizer 117 for inverse quantization of the frame image from the 8 * 8 quantizer 114 and the frame image from the 8 * 8 inverse quantizer 117. 8 * 8 inverse discrete cosine converter 118, and this 8 * 8 inverse discrete cosine converter 118 Adder 119 for adding the frame image from the frame and the generated frame image, a frame memory 120 for storing the frame image from the adder 119, and data of the inputted frame image 16 * 16. By comparing the pixel value between the 16 * 16 blocking unit 123 for dividing into blocks and the current frame image from the 16 * 16 blocking unit 123 and the previous frame image from the frame memory 120. A motion predictor 121 for predicting a motion vector, a motion compensator 122 for applying a motion vector from the motion predictor 121 to a frame image of the frame memory 120 to generate a new frame image, and 8 8 includes a multiplexer 116 which multiplexes the image data of the variable length coding unit 115 and the motion vector from the motion predictor 121.

On the other hand, the resolution of one frame image (the number of horizontal pixels * the number of vertical pixels) is 720 * 480, 1192 * 1080, etc. The types are various, and the movement between the current frame image and the previous frame image in such a frame image In the motion predictor for predicting a vector, one frame image is divided into blocks including 16 * 16 pixels, and one frame image is processed in units of blocks.

The motion predictor 121 compares the pixel value of the current frame image f (t) with the pixel value of the previous frame image f (t-1) and predicts the direction of the motion change. The operation is described in detail below.

FIG. 3 is a block diagram illustrating an example of a block image. In the 16 * 16 blocking unit 123 of FIG. 2, a block including 16 * 16 pixels in one frame image according to MPEG protocol (hereinafter, referred to as “16 * 16 blocks”) is described. A block diagram of the frame image thus blocked is shown in FIG. 3.

FIG. 4 is a partially blocked current frame image diagram of the frame image of FIG. 3. In FIG. 4, a current frame image including eight partial blocks and a total of nine partial blocks around an arbitrary current block B (t) 22 is illustrated in FIG. 4. (f (t)) is shown, and FIG. 5 is a previous frame image diagram including blocks corresponding to the current frame image of FIG. 4, and FIG. 5 corresponds to the current block B (t) 22 of FIG. 4. The previous frame image f (t-1) including eight blocks and a total of nine blocks centering on the block B (t-1) 22 is shown. In FIG. 5, the dotted line is a search window SRW including a plurality of blocks identical to the current block B (t) 22, and the search window SRW is composed of two consecutive frames in an image of approximately 24 frames per second. The decision is made in consideration of the movable range of motion between the images, and in general, an extended range (± block size / 2) is applied to the corresponding block B (t-1) 22.

Referring to FIG. 5, the motion predictor 121 of FIG. 2 compares the current block B (t) 22 of FIG. 4 with a plurality of blocks of the search window SRW of FIG. 5, respectively. Starting from the upper left side shown in FIG. 6A to the lower right side, as shown in FIG. 6B, the matching direction of the electron gun scanning direction of the CRT is found, and thus, the matching vector most similar to the current block is found to predict the motion vector. The algorithm that finds the most similar matching block is called block matching algorithm.

Here, the value to be compared between blocks is the pixel value of each block in the block, and the comparison is a pixel value of the current block of the current frame image and a corresponding value of the corresponding block of the previous frame image, and finding a matching block is a pixel value of the current block. By subtracting the pixel value of the block, finds the block with the least error (difference), calculates the position vector of the matching block centered on the current block, and predicts the motion vector.

On the other hand, the prediction of the motion vector should be made in consideration of two aspects of image quality prevention and fast prediction.

There are various methods for the block matching algorithm, and among these methods, there is a full search method, which is used as an evaluation criterion of other methods, but the image quality of the predicted image is excellent, but one color pixel is used. In order to express, each of the three primary colors (R, G, B) requires 24 bits each for 8 bits. Therefore, processing all the pixel values in the entire area search process requires too much computation. Real-time system implementation will be impossible.

In order to solve this problem, many fast matching algorithms have been proposed, and these fast matching algorithms have been proposed for multi-resolutuon, a method of reducing the search point by unimodal error surface Assumption (UESA). Method, reference search, variable search range (VSR) method using correlation of adjacent motion vectors, a method of detecting calculation in the process of block matching, and the like.

In particular, a representative method of reducing the search point by UESA (Unimodal Error Surface Assumption) is a three step search (TSS) method, which will be described in detail below.

First, FIG. 7 illustrates an example of the size of the search window of FIG. 5. Referring to FIG. 7, the search window SRW has an extended range applied up, down, left, and right with respect to a reference block including 16 * 16 pixels. The search box is larger than the reference block, and the extended range is generally extended to one side of the reference block with a + block size / 2, which will be described below by applying +7 as shown in FIG. do.

8A to 8I are diagrams showing search points in a search box determined by a conventional three-step search method. In the conventional three-step search method, the current block is not searched for matching with all blocks in the search window. As shown in FIG. 8, the center points of the nine blocks to be searched in the search box are displayed as search points '1' to '9' as shown in FIG. 8.

9A and 9C are search point display diagrams for explaining a conventional three-stage search (TSS) method, and the search point display diagrams shown in FIGS. 9A to 9C are first shown in the case of performing a full area search in the search box. The search point that can be the center point of the grid is displayed in the form of a tile, and then nine search points '1 to 9' determined by the three-step search method are displayed.

9A to 9C, a three-stage retrieval method for a conventional motion prediction is as follows.

First, in the process of finding a desired matching block, the pixel value of the current block of the current frame image f (t) and the pixel of the block corresponding to the current block in the search window of the previous frame image f (t-1). The absolute value error sum (hereinafter referred to as 'SAD'), which is the sum of the difference (error) and the value, is mainly used in consideration of the complexity and performance of the calculation. Calculated as

Where Ic () is the pixel value for the block of the current frame image, Ip () is the pixel value for the corresponding block in the search box of the previous frame image, x, y is the coordinate in the search box, and k, l is Coordinates in the block. And n is the size of the matching block.

In addition, using a single-mode error surface assumption (hereinafter, referred to as 'UESA') that an error at each search point increases monotonously as it moves away from the global motion vector, the conventional three-stage retrieval method is 1,2 and The motion vector is determined through a three-step search process. First, in the first-step search, all SADs are calculated for nine search points '1' to '9' in the search window as shown in FIG. 9A. In the next two-step search, when the search point of the minimum value SAD calculated in the first step search, for example, '2' and the SAD is the minimum value, as shown in FIG. 9B, the search point '2' is centered on the '21'. The SADs for all nine search points are calculated again from ˜'29 '. In the three-stage search, in the two-stage search, if the SAD is the minimum value, for example, if the SAD is the minimum value at '22', the SAD for the nine search points around the '22' of the minimum value SAD is calculated. After all the search, the search point having the minimum SAD is determined as the motion vector.

However, the conventional three-step search method performs a search for nine search point modes, respectively, in the 1,2- and 3-step searches. This method includes all nine points in the first step search and one of nine points in the two-step search. 8 points excluding one point calculated in the search, and 8 points excluding one point calculated in the two-step search in the three-step search, that is, 25 points in total (9 points in the first step + eight points in the two-step search) 8 points of +3 step search), which means that 24 bits are required to express one pixel value, and there is a problem that software real-time processing is impossible because the search time is long and the search time is long. .

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and therefore, an object of the present invention is to improve the three-stage retrieval method using a single mode error surface assumption (UESA) to reduce the number of retrieval points, and to improve this. The present invention provides a fast motion prediction method for real-time image compression that suitably applies a half-stop method having a threshold and a partial error to matched errors.

Another object of the present invention is to enable high-speed calculation due to the reduction of the search point while maintaining the predicted picture quality substantially the same as the predicted picture quality by the conventional three-stage retrieval method, so that it can be applied to a real-time system based on software. The present invention provides a fast motion prediction method for compression.

1 is an overall configuration diagram of a general video data transmission and reception system.

FIG. 2 is an internal block diagram of an MPEG source encoder included in the video data transmitter of FIG. 1.

3 illustrates an example of a blocked frame image.

4 is a partially blocked current frame image diagram of the frame image of FIG. 3.

5 is a previous frame image diagram including a block corresponding to the current frame image of FIG. 4.

FIG. 6A is a diagram showing a position where a current block is first searched in a search box, and FIG. 6B is a diagram showing a direction in which a current block is searched in a search box.

FIG. 7 is a diagram illustrating an actual example of the size of the search window of FIG. 5.

8A to 8I are diagrams showing search points in a search box determined by a conventional three-step search method.

9A and 9C are search point display diagrams for explaining a conventional three-step search method.

10 is an overall flowchart showing a fast motion prediction method according to the present invention.

FIG. 11 is a flowchart illustrating a search process of an outer search point in the search process of FIG. 10.

FIG. 12 is a flowchart illustrating a two-step search process centering on a search point of the minimum value SAD in FIG. 11.

FIG. 13 is a flowchart illustrating a search process for an outer search point in the two-step search process of FIG. 12.

FIG. 14 is a flowchart illustrating a three-step search process centering on a search point of the minimum value SAD in FIG. 13.

15A and 15F are search point display diagrams for the flowcharts of FIGS. 10 and 11.

16 is a search point display diagram of the flowcharts of FIGS. 12 and 13.

17A and 17B are search point display diagrams for the flowchart of FIG. 14.

Explanation of symbols on the main parts of the drawings

121: Motion predictor f (t): current frame image

f (t-1): previous frame image BC: current block

SRW: Search box MV: MOTION VECTOR

SAD: Absolute value error sum SADmin: Absolute value error sum minimum

SAD (x): Absolute value error sum for search point 'x'

TH: Match Error Threshold

As a technical means for achieving the object of the present invention described above, the method of the present invention, in order to reduce the search point in the three-stage search method using a single-mode error surface assumption, the center search point and the center search point and up, down, left and right After calculating the SADs for two adjacent outer search points and a total of three points among the four adjacent search points, the search point corresponding to the minimum value SAD among the three SADs is found, and the search point of the minimum value SAD is found. If the search point is any one of the two outer search points, the SAD of the other two outer search points adjacent to the search point in both directions is calculated, and then the two-step search is performed for the search point of the minimum SAD among the three SADs. Perform.

When the minimum value SAD is the center search point, the SADs of two different search points adjacent to each other among the four search points adjacent to the center search point and up, down, left, and right are calculated, and the search corresponding to the minimum value SAD among the three SADs. Find the point, and if the search point of the minimum SAD is one of two outer search points, calculate the SADs of two other outer search points that are bidirectionally adjacent to the search point, and then among the three SADs A two-step search is performed for the search point of the minimum value SAD.

On the other hand, when the minimum value SAD is a central search point, a two-step search is performed centering on the central search point. The two-step search process is the same as the one-step search process described above, and the three-step search process is a conventional three-step search. Is the same as

In addition, by comparing each SAD being calculated in the first and second steps of the search step 1 and SADmin set in advance, if the SAD being calculated is greater than SADmin, the SAD calculation is stopped and proceeds to the next process, one SAD calculation is completed If this SAD is smaller than the matching error threshold (TH), the search point of this SAD is determined as a motion vector.If the SAD calculated is smaller than the initially set SADmin, the SADmin is reset to increase the probability of stopping SAD calculation. Because of this, the fast motion prediction is possible.

EMBODIMENT OF THE INVENTION Hereinafter, this invention is demonstrated with reference to attached drawing.

In a motion predictor of an encoder for compressing image data, a plurality of blocks in a search window SRW of a previous frame image f (t-1) match a block matching the current block of the current frame image f (t). The fast motion prediction method for detecting the motion vector by predicting the motion vector is modified from the existing three-step (first, second and third) search methods in order to reduce the search point.

First, referring to FIG. 15, the first step of the three-step search process will be described. First, the search points '5, 6, and 8' are searched as shown in FIG. 15B. If the minimum value, the SAD of the search point '3, 6, 9' is calculated again and performs a two-step search process around the search point of the minimum value SAD. If the SAD of the search point '8' is the minimum value, as shown in FIG. The SAD of the search points' 7,8,9 'is calculated to perform a two-step search process centering on the search point of the minimum value SAD. On the other hand, if the SAD of the search point' 5 'is the minimum value, the search point' is again shown in FIG. 15D. After calculating the SAD of 2, 4 ', if the SAD of the search point' 4 'is the minimum value among the SADs of the search points' 2, 4, 5', the SAD of the search points' 1, 4, 7 'is again shown in Fig. 15E. Is calculated to perform a two-step search process around the search point of the minimum value SAD, and if the SAD of the search point '2' is the minimum value among the SADs of the search points '2, 4, 5', the search point is again shown in FIG. Search for minimum value SAD by calculating SAD of 1,2,3 ' Mainly carried out a two-phase search process. If the SAD of the search point '5' is the minimum value among the SADs of the search points '2, 4, 5', a two-step search process is performed based on the search point '5'.

In the fast motion prediction method of the present invention, one step of the three steps will be described with reference to FIGS. 10, 11, and 15 as follows.

First, with respect to step (a) of the first stage search, referring to 51, 52 of FIG. 10, a matching error threshold TH is first set, and then a SAD is obtained for any search point in the search box. Is set to SADmin. If it is determined that the set SADmin is smaller than the set matching error threshold value TH, the search point of this SADmin is determined as a motion vector, and if it is not small, the process continues to SAD calculation.

In this case, the preset SADmin is an SAD for any search point set in advance in the search window SRW of the previous frame image f (t-1), which takes into account the center search point of the corresponding block or the previous motion vector. Is set.

Next, referring to 53 of FIG. 10, the nine search points' 1 to 9 shown in FIG. 15 in the first stage search window SRW of the previous frame image f (t-1) by the three stage search method. 'Middle center search point' 5 'and the four outer search points' 2, 4, 6 and 8 'adjacent to each other in the search point' 5 ', the two outer search points shown in Fig. 15A' 6,8 Calculate the SAD for '.

While calculating the SADs of the search points '5', '6' and '8', it is determined whether each calculated SAD is larger than a predetermined SADmin, and if it is not large, the SAD calculation is continued. Stop SAD calculation and proceed to the next step.

Referring to 54 of FIG. 10, it is determined whether there is a SAD smaller than the preset match error threshold TH among the calculated SADs, and if there is a SAD smaller than this match error threshold TH, a search point corresponding to this SAD is determined. If it is determined as a motion vector, otherwise referring to 55 of FIG. 10, the minimum value SAD is found from the calculated SAD.

Secondly, for step (b) during the first step search, referring to 55 of FIG. 10, in the case where the minimum value SAD found in step (a) is any one of the SADs of the two outer search points '6 and 8', For example, when the SAD of the search point '6' is minimum, referring to 61 and 62 of FIG. 11, it is determined whether the SAD of the calculated search point '6' is smaller than the SADmin. Reset to.

Then, the SADs of two outer search points adjacent to the search point of the minimum value SAD are calculated. For example, if the SAD of the search point '6' is the minimum value SAD, as shown in FIGS. 63 and 15B of FIG. SAD of point '3,9' is additionally calculated, and SAD (3,9) being calculated is compared with preset SADmin. If SAD being calculated is bigger than preset SADmin, stop SAD calculation and proceed to next step. Otherwise, SAD calculation is continued.

On the other hand, if the SAD of the search point '8' is the minimum value SAD, it is determined whether the SAD 8 is smaller than the preset SADmin, and if it is small, the SADmin is reset to the SAD 8, as shown in Fig. 15C, the search point. Calculate additional SAD of '7,9', and compare SAD (7,9) being calculated with preset SADmin.If SAD being calculated is bigger than preset SADmin, stop SAD calculation and proceed to next step. If not, continue SAD calculation.

Then, as shown in Fig. 15C, the SAD of the search points '7, 9' is further calculated, and the calculated SAD is larger than the preset SADmin by comparing the calculated SAD (7, 9) with the preset SADmin. If it loses, it stops calculating SAD and proceeds to the next step. Otherwise, SAD calculation continues.

If the SADs of the search points '6' and '8' are not the minimum value SAD, and the SAD of the search point '5' is the minimum value SAD, referring to 56 of FIG. 10, the SAD 5 is smaller than the preset SADmin. If it is small, it resets SADmin to SAD (5), and then nine search points' 1-9 in the first stage search window SRW of the previous frame image f (t-1) by the three stage search method. Two other outer search points '2, 4, 6, and 8' which are adjacent to each other in FIG. The SAD for 4 'is calculated as shown in 57 of FIG.

While calculating the SADs of the search points '5', '2', and '4', it is determined whether each calculated SAD is larger than the predetermined SADmin, and if it is not large, the SAD calculation is continued. Stop the calculation and proceed to the next step.

Then, it is determined whether there is a SAD smaller than the preset match error threshold TH among the calculated SADs, and if there is a SAD smaller than this match error threshold TH, a search point corresponding to the SAD is determined as a motion vector. Otherwise, referring to 59 of FIG. 10, the minimum value SAD is found among the calculated SADs.

In addition, referring to 64 of FIG. 11, it is determined whether the calculated SAD is smaller than the preset matching error threshold value TH. If there is a SAD smaller than the matching error threshold value TH, a search point corresponding to the SAD is found. Determined by the motion vector.

Referring to 65 of FIG. 11, the minimum value SAD is found among the SADs of the three outer search points '3, 6, 9' (or '7, 8, 9') calculated in this way, and reference is made to 65 of FIG. In this case, a two-stage search for a two-stage search box is performed centering on the search point corresponding to the found minimum value SAD.

As such, the minimum SAD obtained in step (b) is the SAD of one of the search points '3, 6, and 9', or the SAD of one of the search points, '7, 8, and 9', where the search is performed. Since the search principle for the point is the same, when the SAD of the search point '3' is the minimum value SAD, a two-step search process has been described with reference to FIGS. 12, 13, and 16, and the rest of the search points are omitted.

Here, in the two-step search process centering on the search point '3', eight search points are displayed as '31, 32,33,34,36,37,38,39 'centered on the search point' 3 '. .

First, referring to 71 and 72 of FIG. 12, when the minimum value SAD is the SAD of the search point '3', it is determined whether the SAD (3) is smaller than the preset SADmin, and when SADmin is smaller, the SADmin is reset to the SAD (3). do.

Referring to 73 of FIG. 12, as shown in FIG. 16A, the center search point among the nine search points 31 through 39 in the two-level search window centered on the search point '3' by the three-step search method. Calculate SAD for two adjacent outer search points '36, 38 'adjacent to each other among 3' and four outer search points '32, 34, 36, 38 'that are adjacent to each other. The minimum value SAD of the determined SAD 3, SAD 36 and SAD 38 is found.

The two-stage search process of the three-stage search according to the present invention is the same as the one-stage search process described above, except that the size of the search box is different from the one-stage search box shown in FIG. 15 and the two-stage search box shown in FIG.

In this way, during the calculation of the SAD, it is determined whether the SAD being calculated is greater than the SADmin, and when it becomes large, the SAD calculation for this search point is stopped, and the process proceeds to the next step.

method.

It is determined whether the calculated SAD is smaller than the preset matching error threshold TH, and if there is an SAD smaller than the matching error threshold TH, a search point corresponding to the SAD is determined as a motion vector.

In the case where the obtained minimum value SAD is either SAD 36 or SAD 38, for example, as shown in Fig. 13, if the minimum value SAD is SAD 36, two adjacent to this search point '36' are shown. SADs for search points '33, 39 'are calculated, whereas if the minimum value SAD is SAD 38, SADs for two search points '37, 39' adjacent to this search point '38' are calculated. The minimum value SAD is found among the SADs of the three outer search points calculated as described above, and a three-step search is performed on the three-step search box centering on the search point corresponding to the found minimum value SAD.

On the other hand, if the minimum value SAD is the SAD of the search point '3', the other two search points adjacent to each other among the four outer search points 32, 34, 36, and 38 adjacent to each other up, down, left, and right in this search point '3' After calculating the SAD for 32,34 ', the minimum SAD of the calculated two SADs of the outer search points '32, 34' and the SAD of the center search point '3' is found.

In the case where the obtained minimum value SAD is one of the SAD 32 and the SAD 34, for example, if the minimum value SAD is the SAD 32, two search points '31, 33 'adjacent to this search point' 32 ' Calculate the SAD for the two search points '31, 37 'adjacent to this search point' 34 ', if the minimum value SAD is SAD 34. The minimum value SAD is found among the SADs of the three outer search points calculated as described above, and the third level search window is performed using the found minimum value SAD as the central search point.

On the other hand, when the obtained minimum value SAD is the SAD of the search point '3', the three-step search is performed on the three-step search box using the search point '3' as the center search point.

Hereinafter, since each of the three step search processes is the same for each search point, in the two step search process when the search point corresponding to the minimum value SAD in the two step search process is '3', the search point of the minimum value SAD is '36'. The case will be described below with reference to Figs. 14 and 17, and the description of the remaining search points is omitted.

Referring to the three-step search process, if the search point of the minimum value SAD is '38', the nine search points including the search point '38' as the central search point and all the search points and the search point '38' in the three-step search box After calculating the SAD for the search point, the search point having the minimum value SAD among the calculated SADs is found and determined as a motion vector.

According to the present invention as described above, the three-stage search method using a single mode error surface assumption (UESA) is improved to reduce the number of search points, and the improved method has a threshold for matching errors and By appropriately applying a half-stop method with partial error, there is an effect that high-speed motion prediction is possible.

In addition, another effect of the present invention is that it is possible to calculate the high speed due to the reduction of the search point while maintaining the predicted picture quality almost the same as the predicted picture quality by the conventional three-stage search method, which can be applied to a software-based real-time system will be.

The above description is only a description of one embodiment of the present invention, the present invention is capable of various changes and modifications within the scope of the configuration.

Claims (32)

Performed by the encoder's motion predictor for compressing the image data, a block matching the current block of the current frame image f (t) is stored in the search window SRW of the previous frame image f (t-1). In the fast motion prediction method for detecting a motion vector by detecting in a block, (a) Among the nine search points' 1 to 9 'in the first search window SRW of the previous frame image f (t-1) by the three-step search method, the center search point' 5 'and this search point' Computing SADs of two outer search points' 6,8 'adjacent to each other among the four outer search points' 2, 4, 6, and 8' adjacent to each other at 5 ', and then finding the minimum SAD among the calculated SADs. ; (b) If the minimum value SAD found in step (a) is one of the SADs of the two outer search points '6,8', after calculating the SADs of the two outer search points adjacent to the search point of this minimum value SAD, Finding a minimum value SAD among SADs of three outer search points, and performing a two-step search for a two-step search box centering on a search point corresponding to the found minimum value SAD; (c) If the minimum value SAD found in step (a) is the SAD of the center search point '5', the four outer search points '2, 4, 6 and 8' adjacent to each other at this search point '5' Calculating SADs for two adjacent outer search points '2,4' and finding a minimum SAD of the calculated SADs of the other two outer search points '2,4' and SADs of the central search point '5'; (d) if the minimum SAD found in step (c) is one of the two SADs of the two outer search points '2,4', after calculating the SADs for the two outer search points adjacent to the search point of the minimum SAD, Finding a minimum value SAD of the SADs of the three outer search points, and performing a two-step search on a two-step search window centering on the search points corresponding to the found minimum value SAD; (e) if the minimum value SAD found in step (c) is the SAD of the central search point '5', performing a two-step search on the two-step search box centered on the search point; Fast motion prediction method for real time image compression. The method of claim 1, wherein the fast motion prediction method (f) obtaining SAD for a preset search point in the search window SRW of the previous frame image f (t-1) and setting the SAD to SADmin; Fast motion prediction method for real-time video compression, characterized in that it further comprises. The method of claim 1 or 2, wherein the fast motion prediction method (g) When the minimum value SAD obtained in the step (e) is the SAD of the search point '5', it is determined whether the SAD (5) is smaller than the SADmin set in the step (f). Resetting to); high speed motion prediction method for real-time video compression, characterized in that it further comprises. The method of claim 1, wherein the fast motion prediction method and (h) setting a matching error threshold (TH). The method of claim 2, wherein step (a) While calculating the SAD for the search point, it is determined whether the SAD being calculated is greater than SADmin, and if it becomes larger, stops calculating the SAD for this search point and proceeds to the next step. Fast Motion Prediction Method. The method of claim 2, wherein step (b) While calculating the SAD for the search point, it is determined whether the SAD being calculated is greater than SADmin, and if it becomes larger, stops calculating the SAD for this search point and proceeds to the next step. Fast Motion Prediction Method. The method of claim 2, wherein step (c) While calculating the SAD for the search point, it is determined whether the SAD being calculated is greater than SADmin, and if it becomes larger, stops calculating the SAD for this search point and proceeds to the next step. Fast Motion Prediction Method. The method of claim 2, wherein step (d) While calculating the SAD for the search point, it is determined whether the SAD being calculated is greater than SADmin, and if it becomes larger, stops calculating the SAD for this search point and proceeds to the next step. Fast Motion Prediction Method. The method of claim 4, wherein step (a) Determining whether the calculated SAD is smaller than the preset matching error threshold TH, and if there is an SAD smaller than the matching error threshold TH, determining a search point corresponding to the SAD as a motion vector; Fast motion prediction method for real-time video compression, characterized in that it comprises a. The method of claim 4, wherein step (b) Determining whether the calculated SAD is smaller than the preset matching error threshold TH, and if there is an SAD smaller than the matching error threshold TH, determining a search point corresponding to the SAD as a motion vector; Fast motion prediction method for real-time video compression, characterized in that it comprises a. The method of claim 4, wherein step (c) Determining whether the calculated SAD is smaller than the preset matching error threshold value TH, and if there is a SAD smaller than the matching error threshold value TH, determining a search point corresponding to the matching SAD value as a motion vector. Fast motion prediction method for real-time video compression. The method of claim 4, wherein step (d) Determining whether the calculated SAD is smaller than the preset matching error threshold value TH, and if there is a SAD smaller than the matching error threshold value TH, determining a search point corresponding to the matching SAD value as a motion vector. Fast motion prediction method for real-time video compression. The method of claim 4, wherein step (e) Determining whether the calculated SAD is smaller than the preset matching error threshold value TH, and if there is a SAD smaller than the matching error threshold value TH, determining a search point corresponding to the matching SAD value as a motion vector. Fast motion prediction method for real-time video compression. The method of claim 1, wherein step (b) (b1) When the minimum value SAD is the SAD of the search point '6', the SAD for the search points '3 and 9' is calculated, and the minimum value SAD among the SADs (3), SAD (9) and SAD (6) is found. step; (b2) if the minimum value SAD is SAD (3), performing a two-stage search for the two-stage search window using the search point '3' as the central search point; (b3) performing a two-stage search for the two-stage search window using the search point '6' as the central search point when the minimum value SAD is SAD (6); (b4) performing a two-stage search for the two-stage search box using the search point '9' as the central search point when the minimum value SAD is SAD (9). Fast motion prediction method. The method of claim 1, wherein step (b) (b1) When the minimum value SAD is the SAD of the search point '8', the SAD for the search points '7, 9' is calculated, and the minimum value SAD among the SADs (7), SAD (9), and SAD (8) is found. step; (b2) if the minimum value SAD is the SAD of the search point '7', performing a two-step search for the two-step search window using the search point '7' as the central search point; (b3) when the minimum value SAD is SAD 8, performing a two-stage search for the two-stage search window using the search point '8' as the central search point; (b4) performing a two-stage search for the two-stage search box using the search point '9' as the central search point when the minimum value SAD is SAD (9). Fast motion prediction method. The method of claim 14 or 15, wherein step (b) (b5) When the minimum value SAD is the SAD of the search point '6', it is determined whether the SAD (6) is smaller than the preset SADmin, and when it is smaller, the SADmin is reset to the SAD (6). If the minimum value SAD is the SAD of the search point '8', determining whether the SAD (8) is smaller than the preset SADmin, and if it is smaller, resetting the SADmin to SAD (8); Fast Motion Prediction Method for Image Compression. The method of claim 14 or 15, wherein step (b1) While calculating the SAD for the search point, it is determined whether the SAD being calculated is greater than SADmin, and if it becomes larger, stops calculating the SAD for this search point and proceeds to the next step. Fast Motion Prediction Method. The method of claim 14 or 15, wherein step (b1) Determining whether the calculated SAD is smaller than the preset matching error threshold TH, and if there is an SAD smaller than the matching error threshold TH, determining a search point corresponding to the SAD as a motion vector; Fast motion prediction method for real-time video compression, characterized in that it comprises a. The method of claim 1, wherein step (d) (d1) When the minimum value SAD is the SAD of the search point '2', the SAD for the search points '1, 3' is calculated, and the minimum value SAD of the SAD (1), SAD (3) and SAD (2) is found. step; (d2) when the minimum value SAD is the SAD of the search point '1', performing a two-step search on the two-step search window using the search point '1' as the central search point; (d3) when the minimum value SAD is SAD (2), performing a two-stage search for the two-stage search window using the search point '2' as the central search point; (d4) performing a two-stage search for the two-stage search box using the search point '3' as the center search point when the minimum value SAD is SAD (3). Fast motion prediction method. The method of claim 1, wherein step (d) (d1) When the minimum value SAD is the SAD of the search point '4', the SAD for the search points '1,7' is calculated, and the minimum value SAD of the SAD (1), SAD (7) and SAD (2) is found. step; (d2) when the minimum value SAD is the SAD of the search point '1', performing a two-step search on the two-step search window using the search point '1' as the central search point; (d3) when the minimum value SAD is the SAD of the search point '4', performing a two-step search for the two-step search window using the search point '4' as the central search point; (d4) performing a two-stage search for the two-stage search box using the search point '7' as the center search point when the minimum value SAD is the SAD of the search point '7'; Fast Motion Prediction Method for Compression. The method of claim 19 or 20, wherein step (d) (d5) When the minimum value SAD is the SAD of the search point '2', it is determined whether the SAD (2) is smaller than the preset SADmin, and when it is smaller, the SADmin is reset to the SAD (2). If the minimum value SAD is the SAD of the search point '4', determining whether the SAD 5 is smaller than the preset SAD min and resetting the SAD min to the SAD 4 if the SAD is smaller. Fast Motion Prediction Method for Image Compression. The method of claim 19 or 20, wherein step (d1) While calculating the SAD for the search point, it is determined whether the SAD being calculated is greater than SADmin, and if it becomes larger, stops calculating the SAD for this search point and proceeds to the next step. Fast Motion Prediction Method. The method of claim 19 or 20, wherein step (d1) Determining whether the calculated SAD is smaller than the preset matching error threshold TH, and if there is an SAD smaller than the matching error threshold TH, determining a search point corresponding to the SAD as a motion vector; Fast motion prediction method for real-time video compression, characterized in that it comprises a. The method of claim 1, wherein the step 2 search process for the step 2 search box of step (e) is (e1) Up, down, left and right from the center search point '5' and the center search point '5' among the nine search points '51 to 59 'in the two-level search window centered on the search point' 5 'by the three-step search method. After calculating the SADs of two adjacent outer search points '56, 58 'among four adjacent outer search points '52, 54, 56 and 58', the calculated SAD (5), SAD (56) and SAD (58) are calculated. Finding a minimum value SAD); (e2) If the minimum value SAD obtained in step (e1) is one of SAD 56 and SAD 68, after calculating the SADs for two search points adjacent to the search point of this minimum value SAD, the three outer edges are calculated. Finding a minimum value SAD among the SADs of the search points, and performing a three-step search for the three-step search window centering on the search point corresponding to the found minimum value SAD; (e3) If the minimum value SAD obtained in step (e2) is the SAD of the center search point '5', the four outer search points '52, 54, 56 and 58 'which are adjacent to each other at this search point' 5 ' Calculating SADs for two other adjacent outer search points '52 and 54 ', and then finding a minimum SAD among the calculated SADs of the other two outer search points '52 and 54' and SAD of the central search point '5'; (e4) If the minimum value SAD obtained in step (e3) is one of the SAD 52 and the SAD 54, after calculating the SADs for two search points adjacent to the search point of this minimum value SAD, the three outer edges are calculated. Finding a minimum value SAD among the SADs of the search points, and performing a three-step search for the three-step search window centering on the search point corresponding to the found minimum value SAD; (e5) performing a three-stage search for the three-stage search box when the minimum value SAD obtained in step (e4) is the SAD of the central search point '5'; Fast motion prediction method for real-time video compression, characterized in that consisting of. The method of claim 24, wherein step (e1) While calculating the SAD for the search point, it is determined whether the SAD being calculated is greater than SADmin, and if it becomes larger, stops calculating the SAD for this search point and proceeds to the next step. Fast Motion Prediction Method. The method of claim 24, wherein step (e1) Determining whether the calculated SAD is smaller than the preset matching error threshold TH, and if there is an SAD smaller than the matching error threshold TH, determining a search point corresponding to the SAD as a motion vector; Fast motion prediction method for real-time video compression, characterized in that it comprises a. The method of claim 14 or 15, wherein the two-step search process for the two-step search box The center search point '3' out of the nine search points '31 to 39 'in the two-level search box centered on the search point' 3 'by the three-step search method, and the four adjacent to the center search point' 3 ' After calculating the SADs for the two outer search points '36, 38 'which are adjacent to each other among the outer search points '32, 34, 36 and 38', the minimum value of the calculated SAD (3), SAD (36) and SAD (38) is calculated. Finding a SAD; If the minimum value SAD obtained in the above step is any one of the SAD 36 and the SAD 38, after calculating the SADs for two search points adjacent to the search point of the minimum value SAD, the SADs of the three outer search points are calculated. Finding a minimum value SAD and performing a three-step search for a three-step search window centering on a search point corresponding to the found minimum value SAD; If the minimum value SAD obtained in the above step is the SAD of the search point '3', two other search points adjacent to each other among the four outer search points 32, 34, 36, and 38 that are adjacent to each other in the search point '3' Calculating a SAD for 32,34 ', and then finding a minimum SAD of the calculated SADs of the other two outer search points '32, 34' and the SADs of the central search point '3'; If the minimum value SAD obtained in the above step is one of the SAD 32 and the SAD 34, after calculating the SADs of two search points adjacent to the search point of the minimum value SAD, the SADs of the three outer search points are calculated. Finding a minimum value SAD, and performing a three-step search for the three-step search window using the found minimum value SAD as a central search point; Performing a three-stage search for the three-stage search window using the search point '3' as the center search point when the minimum value SAD obtained in the step is the SAD of the search point '3'; Fast motion prediction method for real-time video compression, characterized in that consisting of. The method of claim 27, wherein calculating the SAD is When the minimum value SAD is the SAD of the search point '3', determining whether the SAD (3) is smaller than the preset SADmin, and resetting the SADmin to the SAD (3) when it is smaller. Fast Motion Prediction Method for Image Compression. The method of claim 27, wherein calculating the SAD is While calculating the SAD for the search point, it is determined whether the SAD being calculated is greater than SADmin, and if it becomes larger, stops calculating the SAD for this search point and proceeds to the next step. Fast Motion Prediction Method. The method of claim 27, wherein said step Determining whether the calculated SAD is smaller than the preset matching error threshold TH, and if there is an SAD smaller than the matching error threshold TH, determining a search point corresponding to the SAD as a motion vector; Fast motion prediction method for real-time video compression, characterized in that it comprises a. 28. The method of claim 27, wherein the three-stage retrieval of steps b22, b24, and b25 is performed. After calculating the SAD for all search points in the three-step search box with the minimum value SAD as the central search point, A fast motion prediction method for real-time image compression, comprising: finding a search point having the minimum value SAD among the calculated SADs and determining the search point as a motion vector. 25. The method of claim 24, wherein the three-step retrieval process of step (e5) After calculating the SAD for all search points in the three-step search box with the minimum value SAD as the central search point, A fast motion prediction method for real-time image compression, comprising: finding a search point having the minimum value SAD among the calculated SADs and determining the search point as a motion vector.