KR100195734B1 - A motion estimator being extended search windows - Google Patents

Abstract

The present invention relates to a motion estimating apparatus, comprising: a forward motion estimating unit (10) for receiving a macroblock of a current frame and extended search window data of a previous frame by using a plurality of motion estimation chips; A backward motion estimator 20 which receives a macroblock of a current frame and extended search window data of a subsequent frame by using a plurality of motion estimation chips and performs backward motion estimation; A memory 40 for storing the forward motion vector and its absolute error value and the reverse motion vector and its absolute error value for each macroblock; The motion vector and the absolute error value are input from the forward motion estimation unit and the backward motion estimation unit, and the motion vectors and the absolute error values stored in the memory are received and compared with each other to calculate a forward vector and a reverse vector having a minimum error value. The comparison and selection means 30 are effective in determining the search window with a small memory and a relatively simple circuit.

Description

Movement estimator with extended search window

1 is a block diagram showing a motion estimation apparatus according to the present invention.

2 is a diagram illustrating (a) to (c) for explaining a search window extension according to the present invention;

FIG. 3 is a flowchart showing an operation flow processed by the comparison and selection unit shown in FIG.

* Explanation of symbols for main parts of the drawings

10: forward motion estimation unit 20: reverse motion state

30: comparison and selection unit 12, 14, 22, 24: motion estimation chip

40: Memory (SRAM)

The present invention relates to a motion estimator of a video signal, and more particularly, to a motion estimator capable of expanding a search window by connecting a plurality of motion estimators having a predetermined search window.

In general, 'motion estimation' in a video signal processing technique indicates, in vector, how much the pixels of a current frame move compared to a previous frame in a continuous video signal. Instead of estimating a motion vector and transmitting the entire image, a technique of reducing transmission information by transmitting these motion vectors (ie, image compression) is described.

The Block Matching Algorithm (BMA), which is widely used in such motion estimation, assumes that the screen motion is moved horizontally or vertically in parallel, so that the block image of the frame where the motion occurs (that is, the current frame) may be moved. It is a method of estimating which position of the previous frame (that is, the previous frame) is the best match and estimating the position as a motion vector. In this case, 8x8, 16x16 (the number of pixels by the number of pixels by the number of pixels) are mainly used as the size of the block.

Here, in order to find the previous block that is most similar to the reference block of the current frame, the search is performed within a predetermined range around the position of the reference block in the previous frame. Such a range is called a search window. The difference between each candidate block is called a distortion (also called an absolute error (AE)), and indicates the degree of similarity between two blocks.

In addition, comparing all candidate blocks and reference blocks in the search window is called a full search block matching algorithm. f ₁ (x, y), when d f ₂ (x, y) of the current _{frame, f 2 (x, y)} , and f ₁ in the (xa, yb) going to change the a, b f ₁ (xa Find the difference between yb) and f ₂ (x, y), and predict the (a, b) at which the difference is the minimum as a vector to move.

The minimum error between the blocks is called a mean absolute error (MAE) method, and the average absolute error is obtained by the following equation.

In Equation 1, E _abs is the average absolute error value, B is the block size, so if (a, b) is obtained to minimize the E _abs is a motion vector.

On the other hand, the device for performing the motion estimation as described above is implemented in a single chip can be obtained on the market. Since most of these commercial chips provide only a narrow range of search window, a predetermined search window is required to expand the search window. A plurality of motion estimation chips with

In the case of using a plurality of motion estimation chips as described above, a motion vector (MV) and an absolute error value (AE) generated for each motion estimation chip are stored, respectively, and compared at the same time to have a motion vector having the smallest absolute error value (AE). Since each motion tracking chip constitutes a sub window, it should be properly scaled according to the position of the sub window.

However, when configuring an extended search window using a plurality of motion estimation chips as described above, the buffer capacity of the logic to select and compare absolute error (AE) must be large, and a separate memory is required for this purpose. There is a problem that the circuitry becomes complicated due to the increase in the operation. In particular, as the search window widens in the vertical and horizontal directions, the number of motion estimation chips required becomes much more complicated. Therefore, it is desirable to extend the search window using as few motion estimation chips as possible.

Accordingly, the present invention provides a motion estimation apparatus in which a search window is expanded so that a plurality of motion estimation chips can be used in the vertical direction and can be processed in a small operation of less memory by repetitive processing by time division in the horizontal direction. There is a purpose.

In order to achieve the above object, the apparatus of the present invention is a motion estimation apparatus for extending a search window by using a plurality of motion estimation chips having a predetermined search window, wherein the current frame is provided using a plurality of the right-sided estimation chips. Forward motion estimation means for receiving forward motion macro data and expanded search window data of a previous frame; Reverse motion estimation means for receiving a macroblock of a current frame and extended search window data of a subsequent frame by using the plurality of motion estimation chips and performing backward motion estimation; A memory for storing the forward motion vector and its absolute error value and the backward motion vector and its absolute error value for each macroblock; And a forward motion vector and an absolute error value from the forward motion estimation means and the backward motion estimation means, and receive the motion vectors and the absolute error values stored in the memory, and compare them, respectively. And means for comparing and selecting means for calculating a motion vector.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

According to the present invention, a motion estimator that extends a search window by using a multi chip, as illustrated in FIG. 1, has a forward motion estimator 10 and a backward motion estimator 20. The forward motion estimator 10 is composed of a unit 30 and a memory 40, and the forward motion estimator 10 is implemented by two motion estimation chips ME1, 2; 12, 14 having a predetermined search window. Is extended, the backward motion estimator 20 is also implemented by two motion estimation chips ME1,2: 22,24 having a predetermined search window, and the search window is extended in the vertical direction, and the memory 40 Is preferably implemented in SRAM.

Here, the first to fourth motion estimation chips 12, 14, 22, and 24 implementing the forward and reverse motion estimation units are shown in (b) of FIG. 2 with respect to the 16x16 macroblock as shown in (a) of FIG. As shown in Fig. 1, a search window of -8 / + 7 may be provided. Therefore, the forward and backward motion estimator uses two motion estimation chips having a search window as shown in (b) of FIG. 2, and thus -16 / + 15 in the vertical direction as shown in (c) of FIG. It is extended to have a search window of. In this case, a time division method may be used to extend the search window in the horizontal direction, and in the embodiment of the present invention, the search window is repeatedly measured four times while shifting by one block for one slice. Expands to +31

In the embodiment of the present invention, a case in which the motion estimation chip provides a search window of -8 / + 7 for a 16x16 macroblock has been described as an example. However, the technical idea of the present invention is not limited to this embodiment. Applicable for For example, if the motion estimation chip provides a search window of -16 / + 15, using two chips as in the embodiment of the present invention can provide a search window of -32 / + 31 in the vertical direction.

Referring to FIG. 1, the first motion estimation chip 12 of the forward motion estimation unit receives a macro block MB to estimate motion from a modern frame buffer (not shown), and searches for a search window (SW1) from a previous frame buffer (not shown). -U) is inputted, and the candidate block on the search window is compared with the macroblock to output a motion vector having the minimum absolute error and its absolute error value (fwd-mv-U), and a second motion estimation chip of the forward motion estimation unit ( 14) receives a macro block MB for motion estimation from a current frame buffer (not shown), receives a search window (SW1-D) from a previous frame buffer (not shown), and compares the candidate blocks on the search window with the macro blocks. Outputs the motion vector with the minimum absolute error and its absolute error value (fwd-mv-D).

At this time, the search window input to the first motion estimation chip 12 is the upper portion SW1-U of the -16 / + 15 search window formed on the previous frame, as shown in FIG. The search window input to the two motion estimation chip 14 is the lower portion SW1 -D of the -16 / + 15 search window formed on the previous frame.

In addition, the third motion estimation chip 22 of the backward motion estimation unit receives a macro block MB to estimate motion from a current frame buffer (not shown), and inputs a search window SW2-U from a frame buffer not shown. And compares the candidate block on the search window with the macroblock and outputs a motion vector having the minimum absolute error and its absolute error value (bwd-mv-U), and the fourth motion estimation chip 24 of the backward motion estimation unit is not shown. After receiving the macro block (MB) to estimate the motion from the current frame buffer, the search window (SW2-D) is input from the frame buffer (not shown), and compares the candidate block and the macroblock on the search window with the minimum absolute error. Outputs the motion vector and its absolute error value (bwd-mv-D).

At this time, the search window input to the third motion estimation chip 22 is the upper portion SW2-U of the -16 / + 15 search window formed on the frame, as shown in FIG. The search window input to the four motion estimation chip 24 is the lower portion (SW2-D) of the -16 / + 15 search window formed on the next frame.

In addition, the motion estimation technique described above is implemented as a semiconductor chip, which is ST13220 manufactured by SGS-THOMSON, which is a 16x16 macro as shown in (a) and (b) of FIG. A search window of -8 to +7 in the horizontal direction and -8 to +7 in the vertical direction can be provided for the block.

Meanwhile, in FIG. 1, the comparison and selection unit 30 performs the same operation as the flowchart shown in FIG. 3, which is estimated from the upper window of the previous frame from the first motion estimation chip 12 of the forward motion estimation unit. The motion vector and the absolute error value (fwd-mv-) estimated from the lower window of the previous frame are received from the second motion estimation chip 14 of the forward motion estimation unit by receiving one motion vector and the absolute error value (fwd-mv-U). D) and a motion vector and an absolute error value (bwd-mv-U) measured from the upper window of the next frame from the third motion estimation chip 22 of the backward motion estimation unit. The motion vector and the absolute error value (bwd-mv-D) estimated from the lower window of the subsequent frame are input from the motion estimation chip 24, and these absolute error values are compared to the forward motion vector having the small absolute error value. And stores the motion vector direction and the absolute error value (MV) in the memory (40). At this time, the previously processed forward and backward motion vectors and absolute error values (fwd-min-mv bwd-min-mv) are stored in the memory 40, and the absolute errors are minimum as shown in Equations 2 and 3 below. The in forward motion vector New-fwd-MV and the backward motion vector New-bwd-MV are obtained and stored in the memory 40.

That is, during the processing of one slice four times, as in Equation 2, the absolute error value (fwd-mv-U) of the motion vector input from the first motion estimation chip 12 and the second motion estimation chip ( Double absolute error by comparing the absolute error (fwd-mv-D) of the motion vector inputted from 14) and the absolute error (fwd-min-mv) of the forward minimum motion vector of the previous cycle stored in the memory 40. The motion vector having the smallest value becomes the new forward motion vector (New-fwd-MV) and is input from the absolute error value (bwd-mv-U) and the fourth motion estimation chip of the motion vector input from the third motion estimation chip. The absolute error (bwd-mv-D) of the motion vector is compared with the absolute error (bwd-min-mv) of the backward minimum motion vector of the previous cycle stored in the memory 40. The vector becomes a new backward motion vector (New-bwd-MV).

Here, since the motion vector and the corresponding absolute error value are treated as a pair, they are designated and used in the same manner, but compared with each other are absolute error values of the corresponding vector, and finally selected and outputted are motion vectors having a minimum absolute error value. When the forward or backward new motion vector (New-fwd-MV. New-bwd-MV) obtained by Equations 2 and 3 is stored in the memory 40, the minimum motion vector (fwd-mon-mv, bwd) of the previous cycle is stored. -min-mv), and this process is performed four times, and finally the motion vector and the absolute error of the macroblock are stored.

Next, the operation of the present invention will be described in more detail with reference to FIG. 3.

In a preferred embodiment of the present invention, a picture of one frame consists of 34 slices, one slice consists of 60 macroblocks, and four motion estimations per slice are required when encoding according to the present invention. . That is, in the embodiment of the present invention, the search window expansion in the vertical direction is performed by using a plurality of motion estimation chips, and the search window expansion in the horizontal direction is performed by repetitive processing by time division.

In FIG. 3, first, the slice count value is initialized to 0 (slice-count = 0: 101), and then the slice count value is increased by one (slice-court ++: 102) to process the slice, then one. In order to estimate 4 motions for the slice of, the repetition count is initialized to 0 (rep-count = 0: 103), and the repetition count value is increased by 1 (rep-count ++: 104), and then the count value of the macroblock is 0. After initializing with (MB-count = 1: 105), increase the count value of the macroblock (MB-count ++ 106). Subsequently, a pair of forward motion vectors (fwd-mv-U and fwd-mv-D) are input from the first motion estimation chip 12 and the second motion estimation chip 14 of the forward motion estimator, A pair of backward motion vectors (bwd-mv-U and bwd-mv-D) are received from the third motion estimation chip 22 and the fourth motion estimation chip 24 and scaled with reference to the repetition count (rep-count). At the same time, 108 reads the minimum forward motion vector fwd-min-mv and the backward motion vector bwd-mon-mv from the memory 40 from the memory 40 at the same time.

Subsequently, a new forward minimum motion vector (New-fwd-MV) is obtained according to Equation 2 described above, and a new backward motion vector (New-bwd-MV) is obtained according to Equation 3, and these are stored in a memory (SRAM: 40). Store (109, 110). Subsequently, it is determined whether the count value (MB-count) of the macroblock is 60 or more (MB-count60?), And if it is less than or equal to 111, the process returns to step 106 and the process (106, 107, 108, 109) for the next macroblock. , 110, 111) and if the count value of the macroblock is 60 or more, it is determined whether the repetition count (rep-count) is 4 or less (112). Repeat (104, 105, 106, 107, 108, 110, 111, 112). If the repetition count (rep-count) exceeds 4, it is determined whether the slice count value is 34 or more (113). If the value is less than 34, the above steps (102 to 113) are repeated for the next slice of the same picture. If it exceeds 34, the next picture (frame) is input and restarted from the beginning (114).

As described above, the search window is expanded in the vertical direction by using a plurality of motion chips that provide a predetermined search window, and the search window expansion in the horizontal direction is repeated by a predetermined number of times for the same slice using a time division technique. In this relatively simple circuit, the search window can be extended.

For example, a motion estimation chip having a search window of -8 / + 7 in the vertical, horizontal, and direction directions of at least 8 motions when the search window is extended by + 16 / -15 in the vertical direction and -32 / + 31 in the horizontal direction. Estimation chips are required (at least 16 in consideration of the reverse motion main branch), and complex computational circuits are required to store and compare the motion vectors output from these eight motion estimation chips with absolute error values. It can be implemented using a motion estimation chip (four considering the backward motion estimation).

Claims (4)

A motion estimator for expanding a search window by using a plurality of motion estimation chips having a predetermined search window, wherein the plurality of motion estimation chips are used to receive a macroblock of a current frame and extended search window data of a previous frame. Forward motion estimation means (10) for estimating forward motion; Reverse motion estimating means (20) for estimating backward motion by receiving a macroblem of a current frame and extended search window data of a subsequent frame using a plurality of motion estimation chips; A memory 40 for storing the forward motion vector and its absolute error value and the reverse motion vector and its absolute error value for each macroblock; And receiving a motion vector and an absolute error value from the forward motion estimating means 10 and the backward motion estimating means 20, receiving a motion vector and an absolute error value stored in the memory 40, and comparing them with each other to obtain a minimum absolute value. 2. A motion estimation apparatus with an extended search window, comprising a comparison and selection means (30) for calculating forward and backward motion vectors with error values. The motion estimation apparatus of claim 1, wherein the motion estimation apparatus extends the search window in a horizontal direction by repeating motion estimation operations per slice four times in an image frame. The method of claim 1, wherein the forward motion estimating means (10) comprises: a first motion estimation chip (12) for motion estimation by receiving the macro block of the current frame and the search window on the previous frame, and the other on the macro frame and the previous frame; And a second motion estimating chip (14) configured to receive the search window and estimate the motion to expand the search window in a vertical direction. The method of claim 1, wherein the backward motion estimating means (20) comprises a third motion estimation chip (22) for motion estimation by receiving a macro block of a current frame and a search window on a subsequent frame; And a fourth motion estimating chip (24) configured to receive the search window and estimate the motion to extend the search window in a vertical direction.