KR100928272B1 - Motion estimation method and apparatus for video coding - Google Patents

Abstract

The present invention provides a CME macroblock buffer for storing only odd rows in a macroblock of a current frame, a CME reference region buffer for storing only odd rows of luminance signals of a reference frame corresponding to a macroblock of a current frame, and the CME macroblock buffer and the A motion estimation apparatus may be provided in a video encoding including a two-pixel estimator that performs two-pixel estimation corresponding to a value stored in a CME reference region buffer.

H.264, motion estimation

Description

Apparatus and Method for motion estimation for moving picture coding

The present invention relates to a motion estimation method and apparatus for video encoding. In particular, the present invention relates to a motion estimation method and apparatus for efficiently operating a processor in a motion estimation method performed for H.264 video encoding.

The present invention is derived from the research conducted as part of the IT growth engine technology development project of the Ministry of Information and Communication and the Ministry of Information and Communication Research and Development. [Task Management Number: 2006-S-026-01, Title: MPCore Platform-Based Mobile Multimedia SoC]

ITU-T, an international video coding standards organization, and ISO together formed JVT and released H.264. Like the existing video coding schemes, the H.264 standard constructs an estimation signal by estimating motion from a previous frame that has been previously encoded, and uses motion compensation and DCT conversion methods for estimation error signals. Mix it up. In order to improve the compression performance, the H.264 standard performs motion estimation up to 4x4 block units, which are smaller block sizes than conventional methods, and performs motion compensation up to 1/4 pixel units. Based on these technologies, the H.264 video coding standard has twice the compression performance compared to the MPEG-4 ASP, which is the existing video coding standard. Therefore, its application is expected as a core technology to be used for various multimedia communication services, and researches for improving the performance and technology for compressing video data using the H.264 video codec have been actively conducted.

However, the H.264 codec standard coding technique has difficulty in applying to real-time applications because of the high memory bandwidth and computational complexity required for implementation. In particular, efficient motion interpolation and motion estimation are required because motion estimation is performed in units of 1/4 pixels having higher complexity than conventional 1/2 pixel units. In addition, the use of complex methods has required the efficient use of memory at each estimation step.

An object of the present invention is to provide a motion estimation method and apparatus for video encoding.

Another object of the present invention is to provide a motion estimation method and apparatus for video encoding, which can efficiently use a memory when performing motion estimation during H.264 video encoding.

In order to achieve the above objects, according to an aspect of the present invention, a Coarse Mode Estimation (CME) macroblock buffer for storing only odd rows in a macroblock of a current frame, the luminance signal of the reference frame corresponding to the macroblock of the current frame A motion estimation apparatus may be provided in a video encoding including a CME reference region buffer for storing only odd rows of data and a two-pixel estimator for performing 2-pixel estimation corresponding to values stored in the CME macro block buffer and the CME reference region buffer. have.

In a preferred embodiment, the CME macroblock buffer may further include a first CME macroblock buffer that stores only odd rows and a second CME macroblock buffer that stores only even columns. The CME reference region buffer may further include a first CME reference region buffer storing only odd columns and a second CME reference region buffer storing only even columns.

The apparatus may further include a CME reference region buffer controller which receives a value stored in the CME reference region buffer and a motion vector value estimated by the two pixel estimator and transmits the value corresponding to the FME reference region buffer. The even row in the macroblock of the current frame may be included in the FME macroblock buffer of the quarter motion controller connected to the next stage of the motion estimation device. The even row of the luminance signal of the reference frame corresponding to the macro block of the current frame may be included in the FME reference region buffer of the quarter motion adjustment device connected to the next stage of the motion estimation device. .

According to another aspect of the present invention, storing only odd rows in the macro block of the current frame, storing only odd rows of the luminance signal of the reference frame corresponding to the macro block of the current frame and the stored current frame A motion estimation method may be provided in a video encoding comprising estimating a motion of a two pixel unit corresponding to an odd row of a macroblock odd row and an odd row of the luminance signal of the reference frame.

In a preferred embodiment, storing only the even numbered rows in the macroblock of the current frame, storing only the even rows of the luminance signal of the reference frame corresponding to the macroblock of the current frame and the stored current frame macroblock The method may further include estimating a quarter-pixel unit motion corresponding to the even row, the odd row, the odd row of the luminance signal of the reference frame, and the even row. In addition, storing only odd rows in the macroblock of the current frame may further include storing odd columns among the odd rows and even columns among the odd rows in physically different memories. The storing of only odd rows of the luminance signal of the reference frame corresponding to the macro block of the current frame may further include storing odd columns of the odd rows and even columns of the odd rows in physically different memories. can do. The method may further include receiving a value included in an odd row of the luminance signal of the reference frame and the two pixel motion estimated motion vector value and converting the value into a value corresponding to an FME reference region buffer.

According to the present invention, a motion estimation method and apparatus for video encoding can be provided.

In addition, the present invention can provide a motion estimation method and apparatus for video encoding that can efficiently use the memory when motion estimation when H.264 video encoding.

Hereinafter, a motion estimation method and apparatus for video encoding according to the present invention will be described in detail with reference to the accompanying drawings.

1 is a block diagram illustrating an overall H.264 encoding apparatus to which the present invention is applied.

Referring to FIG. 1, an H.264 encoding apparatus to which the present invention is applied includes a quantization unit 101, an inverse quantization unit 103, an entropy encoding unit 105, a block release unit 107, a reference frame reproduction unit 109, The motion compensator 111, the motion estimator 113, and the intra predictor 115 are included.

The quantization unit 101 is responsible for quantizing the macro block input from the outside according to a predetermined method. A representative image transformation technique quantized by the quantization unit 101 is DCT (Discrete Cosine Transform).

The inverse quantization unit 103 receives inversely quantized image data after conversion from the quantization unit 101 and performs inverse quantization and inverse transformation.

The entropy encoder 105 is responsible for generating an output bit stream by entropy encoding a final macroblock in which the last h.264 encoding operation is completed.

The block releasing unit 107 serves to release a block of the dequantized macro block to provide a natural image.

The reference frame reproducing unit 109 is responsible for generating and storing a plurality of reference frame images by using the image received from the block release unit 107. These reference frames are a plurality of frame images used as reference screens for motion estimation and motion compensation.

The motion compensator 111 (MC) is responsible for compensating the macro block based on the reference frame stored in the reference frame reproducing unit 109 and the motion vector and the cost value estimated by the motion estimator 113. do.

The motion estimation unit 113 is responsible for estimating the motion of the current macro block by using the reference frame stored in the reference frame reproducing unit 109 and calculating a motion vector and a cost thereof. Unlike the conventional MPEG method, such a motion estimator estimates motion in units of 1/4 pixels, which is very complicated and complicated. In particular, the motion estimator 113 may be classified into a coarse mode estimator (CME) and a fine motion estimator (FME), and a method of estimating such motion is an essential part of the present invention. As will be described in more detail in the following drawings.

The intra prediction unit 115 performs a role of performing prediction encoding, that is, intra prediction, in a macro block by using the reference frame stored in the reference frame reproducing unit 109.

2 is a view for explaining the structure of the motion estimation unit according to the present invention.

Referring to FIG. 2, the motion estimation unit largely includes a CME unit 220 and an FME unit 230. The CME unit includes a CME reference area buffer 221, a CME macro block buffer 223, and a two pixel estimation unit 225. The FME unit 230 includes an FME reference area buffer 233, an FME macro block buffer 231, and a quarter pixel estimator 237.

The CME reference area buffer 221 stores only odd rows of luminance signals of the reference frame corresponding to the macro block of the current frame from the memory 210 that stores both the reference frame and the current frame included in the reference frame reproducing unit. In charge. The conventional CME reference region buffer stores all the luminance signals of the current reference frame. However, in the present invention, only odd rows can be stored, thereby significantly reducing the size of the memory and reducing the repeated transmission time from 210.

The CME macro block buffer 223 stores only odd rows of macro blocks of the current frame from the memory 210 that stores both the reference frame and the current frame included in the reference frame reproducing unit. Although the conventional CME macroblock buffer stores all the macroblock signals of the current reference frame, in the present invention, only odd rows can be stored, thereby significantly reducing the size of the memory. That is, if 16x16 macroblock signals of the current frame stored in the existing CME macroblock buffer are 8x16 in the present invention.

The two pixel estimator 225 estimates a two pixel unit motion vector using the values stored in the CME macro block buffer 223 and the CME reference region buffer 221. This estimation method will be described with reference to FIG. 3.

The FME reference area buffer 233 is a buffer that stores, in units of one pixel, the macro block of the reference frame corresponding to the macro block of the current frame to be motion compensated.

The FME macro block buffer 231 is a buffer that stores only the value of the macro block that is currently calculated among the pixel values of the current frame existing in the memory 210.

In particular, the FME reference area buffer 233 and the FME macro block buffer 231 store only even rows of the current frame macro block in the present invention. This is because the odd rows stored in the CME reference region buffer 221 and the CME macro block buffer 223 described above can be used as they are.

The quarter pixel estimator 237 uses the value stored in the FME reference area buffer 233 and the FME macro block buffer 231 and the value stored in the CME unit to determine one pixel unit, 1/2 pixel unit and 1 /. It is responsible for estimating the motion vector in units of 4 pixels.

3 is a view for explaining luminance motion estimation to which the present invention is applied.

Referring to FIG. 3, the luminance motion estimation applied in the present invention is a difference value Δx, Δy from a macroblock 300 of a frame currently input and a macroblock 310 of a reference frame set from a plurality of frames in the past. ) To calculate the motion vector and its cost. When such a motion estimation is possible, by transmitting only a frame that compensates for a motion estimation value and a motion vector indicating a difference value of the motion, a continuous screen can be configured and the size of a video signal can be greatly reduced.

In this case, the motion estimation calculates a cost by obtaining a difference between the absolute values of the pixels of the current frame macro block 300 and the reference frame macro block 310, and calculates a cost of the reference frame. The pixel point Δx and Δy having the lowest cost may be defined as a motion vector.

These motion vectors and cost values

Motion Vector = (Δx, Δy) MinCost

MINCOST = MIN {Cost (Δx, Δy)}

Cost (Δx, Δy) =

It can be calculated as Where Luma represents luminance.

4 is a diagram illustrating a method of storing a current frame macro block in a CME macro block buffer according to the present invention.

Referring to FIG. 4, assuming that the current frame macro block 400 has 16 × 16 blocks, the CME macro block buffer 411 included in the CME unit includes a frame macro block corresponding to the solid arrow in the figure. Only odd rows of 400 are stored.

In addition, an even row of the frame macroblock 400 not stored in the CME macroblock buffer 411 is stored in the FME macroblock buffer 413 included in the FME block, which is the next stage of the CME block.

When the current frame macro block 400 is alternately stored in each buffer as described above, the two-pixel estimation 421 performed by the CME unit is oddly horizontally and vertically in the macro block among values included in the current frame macro block 400. Since the estimation is performed on only the values included in the column, the estimation is not performed using all the values included in the current frame macro block 400, but only the value included in the odd column among the values included in the macro block. It is targeted. In other words, only the value of 8x8, which is a value included in the odd column, is actually used among the 16x16 macroblocks.

Therefore, since the CME macroblock buffer 411 only needs to store the macroblock value of the current frame for performing the 2-pixel estimation included in the CME, only the value corresponding to the odd row needs to be stored.

Meanwhile, in the FME performed after the CME, motion estimation is performed up to a quarter pixel unit. In this case, all values of the macro block 400 of the current frame are used from the quarter pixel estimation 423.

Therefore, the FME macro block buffer 413 stores values of the remaining even rows of the macro block not stored in the CME macro block buffer 411.

Then, when performing the 1/4 pixel estimation 423 in the FME unit, not only the even row values included in the FME macro block buffer 413 but also the odd row values included in the CME macro block buffer 411. All will be used.

5 is a diagram illustrating a memory storing method of a CME macroblock buffer according to the present invention.

Referring to FIG. 5, the CME macroblock buffer 500 according to the present invention stores an odd row 501 of the current frame macroblock buffer. In this case, since it operates in a general 32-bit environment and each block value has an 8-bit value, a block read in one process will be 4 blocks.

On the other hand, since the value required for the two-pixel estimation of the present invention requires the value of the block cell corresponding to the odd row and the odd column, the block column of the odd column indicated by A in this figure among the four blocks actually read in one process Only values corresponding to are needed.

Accordingly, the CME macroblock buffer of the present invention stores only the first CME macroblock buffer 510 and the even-numbered block 513 in the current frame macroblock 501 including only odd-numbered rows. The second CME macroblock buffer 520 is physically separated and configured. In this case, the actual buffer required for performing the 2-pixel estimation is only the first CME macro block buffer 510. All four block values read in one process are actually used for motion estimation, and thus the number of reads is increased. Can be reduced.

6 is a diagram illustrating a method of storing a current frame reference block in a CME reference region buffer according to the present invention.

Referring to FIG. 6, only odd rows of the frame reference block 600 corresponding to the solid arrow in the figure are stored in the CME reference region buffer 611 included in the CME unit of the present frame reference block 600. .

In addition, an even row of the frame reference block 600 not stored in the CME reference region buffer 611 is stored in the FME reference region buffer 613 included in the FME unit, which is the next stage of the CME unit.

When the current frame reference block 600 is alternately stored in each buffer as described above, the two-pixel estimation 621 performed by the CME unit is an odd number horizontally and vertically in the reference block among values included in the current frame reference block 600. Since the estimation is performed on only the values included in the column, the estimation is not performed using all the values included in the current frame reference block 600, but only the values included in the odd number among the values included in the reference block. It is done.

Therefore, since the CME reference region buffer 611 needs to store only the reference block value of the current frame for performing only 2 pixel estimation included in the CME, only the value corresponding to the odd row needs to be stored.

On the other hand, the FME unit performed after the CME unit performs the unit of 1/4 pixel estimation, wherein all values of the reference block 600 of the current frame are used from the 1/4 pixel estimation 623.

Therefore, the FME reference area buffer 613 stores values of the remaining even rows of the reference block not stored in the CME reference area buffer 611.

Then, when performing the one-pixel estimation 623 in the FME unit, not only the values of the even rows included in the FME reference region buffer 613 but also the values of the odd rows included in the CME reference region buffer 611 are used. Done.

In this case, the block value of the reference frame included in the FME reference region buffer 613 has a value extended by -4/4 on the x axis and -4/4 on the y axis based on the CME motion vector 617. Therefore, the value contained in the CME reference region buffer 611 cannot be used as it is, and an FME controller 615 is added to the CME reference region buffer 611 to add the CME motion vector 617 value obtained from the CME unit. The FME controller 615 reads as much as 12x24 at the point of -4 on the x-axis and -4 on the y-axis based on the current macro block.

When the FME control unit 615 performs the 1/4 pixel estimation 623 in the FME step, both the odd row value and the even row value of the current frame reference block can be obtained.

7 is a diagram illustrating a memory storing method of a CME reference region buffer according to the present invention.

Referring to FIG. 7, the CME reference region buffer 700 according to the present invention stores an odd row 701 of the current frame reference region buffer. In this case, since it operates in a general 32-bit environment and each block value is 8 bits in size, the block read in one process will be 4 blocks.

On the other hand, since the value required for the 2-pixel estimation of the present invention requires a block value corresponding to an odd row and an odd column, only a block value of an odd column indicated by A in this drawing is actually required among four blocks read in one process. do.

Accordingly, the CME reference region buffer of the present invention stores only the first CME reference region buffer 710 and the even-numbered block 713 in the current frame macro block 701 including only odd-numbered rows. The second CME reference region buffer 720 is physically separated and configured. In this case, the actual buffer required for performing the 2-pixel estimation is only the first CME reference region buffer 710, and all four block values read in one process are substantially used for motion estimation, thereby reducing the number of reads. Can be reduced.

The present invention is not limited to the above embodiments, and many variations are possible by those skilled in the art within the spirit of the present invention.

BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 is a block diagram illustrating an overall H.264 encoding apparatus to which the present invention is applied.

2 is a diagram for explaining the structure of a motion estimation unit according to the present invention;

3 is a view for explaining luminance motion estimation to which the present invention is applied;

4 is a diagram illustrating a method for storing a current frame macro block in a CME macro block buffer according to the present invention.

5 is a diagram illustrating a memory storage method of a CME macroblock buffer according to the present invention.

6 illustrates a method of storing a current frame reference block in a CME reference region buffer according to the present invention.

7 is a diagram illustrating a memory storage method of a CME reference region buffer according to the present invention.

<Explanation of symbols for the main parts of the drawings>

220: CME part 230: FME part

221: CME reference area buffer 223: CME macro block buffer

225: 2 pixel estimator 233: FME reference region buffer

231: FME macroblock buffer 237: 1/4 pixel estimation unit

Claims (11)

Coarse Mode Estimation (CME) macro block buffer for storing only odd rows of macro blocks of a current frame, wherein the CME macro block buffer includes only a first CME macro block buffer for storing only odd columns of odd rows and even columns of odd rows. Further comprising a second CME macro block buffer for storing; A CME reference region buffer storing only odd rows of luminance signals of the reference frame corresponding to the macro block of the current frame; A two-pixel estimator for performing two-pixel estimation corresponding to values stored in the CME macro block buffer and the CME reference region buffer Motion estimation apparatus in the video encoding comprising a. delete The method of claim 1, The CME reference region buffer further comprises a first CME reference region buffer storing only odd columns of the odd rows and a second CME reference region buffer storing only even columns of the odd rows. The method of claim 1, And a CME reference region buffer control unit which receives a value stored in the CME reference region buffer and a motion vector value estimated by the two pixel estimator, and transmits the value to a value corresponding to the FME reference region buffer. The method of claim 1, The even row of the macroblock of the current frame is included in the FME macroblock buffer of the quarter motion control device connected to the next stage of the motion estimation device. The method of claim 1, The even row of the luminance signal of the reference frame corresponding to the macro block of the current frame is included in the FME reference region buffer of the quarter motion controller connected to the next stage of the motion estimation apparatus. Estimation device. Storing only odd rows of macro blocks of the current frame, wherein odd and even columns of odd rows of the macro blocks of the current frame are stored in physically different memories; Storing only odd rows of luminance signals of the reference frame corresponding to the macro block of the current frame; and Estimating a two-pixel unit motion corresponding to the stored odd row of the current frame macro block and the odd row of the luminance signal of the reference frame. Motion estimation method in the video encoding comprising a. The method of claim 7, wherein Storing only even rows of the macroblock of the current frame; Storing only even rows of luminance signals of a reference frame corresponding to the macro block of the current frame; and Estimating the motion of the unit of 1/4 pixel corresponding to the even row, the odd row, and the odd row and even row of the luminance signal of the reference frame of the stored current frame macro block. Motion estimation method in the video encoding further comprising. delete The method of claim 7, wherein Storing only odd rows of the luminance signal of the reference frame corresponding to the macro block of the current frame And storing the odd columns of the odd rows and the even columns of the odd rows in physically different memories. The method of claim 7, wherein And receiving a value included in an odd row of the luminance signal of the reference frame and the two-pixel motion estimated motion vector value and converting the value into a value corresponding to an FME reference region buffer.

Images