KR100757830B1 - Method for compressing moving picture using 1/4 pixel motion vector - Google Patents

Abstract

The present invention relates to video compression and reconstruction. Particularly, when a video is compressed through motion prediction using a motion vector, the motion image is expanded to a quarter-pixel to obtain a motion vector, thereby compressing the video. The present invention relates to a video compression method using a 1/4 pixel motion vector.

In the conventional conventional motion vector detection method, a motion vector in units of one pixel is obtained, and then half-pixels are obtained using interpolation between a macroblock corresponding to the pixel block and a macroblock corresponding to the adjacent pixel, that is, a simple average value between the two macroblocks. Find the motion vector up to half-pixel.

In the present invention, interpolation between a macroblock for a motion vector in units of 1 pixel and a macroblock corresponding to a pixel adjacent thereto is calculated using a weight according to a position, not a simple average value, to calculate a unit of a motion vector up to at least 1/4 pixel units. As a result, the image compression efficiency is increased, so that after the image decoding is completed, more improved image quality can be provided.

Motion vector, 1/4 pixel

Description

Method for compressing moving picture using 1/4 pixel motion vector}

1 is a block diagram illustrating a video compression encoding apparatus using a general motion vector.

2 is a block diagram illustrating a video decoding apparatus using a general motion vector.

3 is a diagram illustrating a process of obtaining a general motion prediction and a motion vector.

4 is a diagram for describing a process of obtaining a conventional 1/2 pixel motion vector.

5 and 6 illustrate an embodiment for explaining a process of calculating a 1/4 pixel motion vector according to the present invention.

FIG. 7 illustrates a motion compensation method based on a 1/4 pixel motion vector in the present invention.

The present invention relates to video compression and reconstruction. Particularly, when a video is compressed through motion prediction using a motion vector, the motion vector is extended to a quarter-pixel to obtain a motion vector, and the video is compressed accordingly. The present invention relates to a video compression method using a 1/4 pixel motion vector.

Since the standard for compressing and decompressing video (ISO / IEC 13818-2) has become popular, many techniques using a Motion Picture Expert Group (MPEG) have been widely used in various fields, and the configuration thereof is illustrated in FIGS. 1 and 2.

First, FIG. 1 is a block diagram schematically illustrating a configuration of a video compression encoding apparatus. An input image stored in a field or frame unit in a field / frame storage unit 1 is subtracted from a DCT unit 3 through a subtractor 2. (Discret Cosine Transform) is processed and arranged in units of frequencies by field or frame block, and then the quantization unit 4 performs quantization.

The image quantized by the quantization unit 4 is coded by the variable length coding unit (VLC) 5.

The bit streams compressed in this way are temporarily stored according to the control and transmission rate of the buffer and then output through the buffer 6.

At this time, the subtractor 2 obtains the difference between the input image input from the field / frame 1 and the image input from the motion compensator 12 and performs the DCT processing in the DCT unit 3. When the image output from the field / frame storage unit 1 is an I picture, since there is no motion compensation image from the motion compensator 12 (MC), it is encoded and output as it is. do.

However, in the case of a P picture or a B picture, the subtractor 2 subtracts the image inputted from the field / frame storage unit 1 and the motion predicted and compensated image inputted from the motion compensator 12. To run the DCT.

As described above, in order to output the motion compensated image from the motion compensator 12, an inverse quantization process is performed on the image quantized by the quantizer 4 through the inverse quantizer 7. In addition, the inverse DCT is executed again through the inverse DCT unit 8 to restore the current image.

The reconstructed image corresponds to the image value output through the subtractor 2.

The reconstructed image is stored in the memory 10 and used as a reference for a subsequent image to be used for motion estimation of the next image.

The motion estimator 11 (ME) predicts the motion of the next image to be encoded with respect to the current image thus restored.

By performing block matching between the current image and the image to be motion predicted, a motion vector having the best matching is obtained.

The motion compensation unit 12 outputs a motion compensated image according to the motion vector value thus obtained.

The motion-compensated image is output to the subtractor 2, and the image which has been delayed for a predetermined time through the delay unit 13 and subjected to the motion quantization step is subjected to inverse quantization again as described above to add up to the summer 9 Is input to the summer 9 in synchronization, and the summed values are stored in the memory 10 and used for motion prediction of the next image, as described above.

In other words, the motion-compensated image, which is compressed and then reconstructed, that is, the motion-compensated image that is output through the delay unit 13, is reconstructed to be used to predict the motion of the next image.

The video that is compressed and output by the same process as described above is decoded by a decoder, and FIG. 2 shows the configuration thereof.

When the compressed-coded video is input in the form of a bit stream as described above, the variable-length decoding unit 21 processes the macroblock in units of macroblocks to perform variable length decoding, and performs inverse scan. The restored values for the ground macro blocks are generated in the form of a two-dimensional array.

As described above, the decoded image is dequantized through the inverse quantization unit 22 and the inverse DCT is executed through the inverse DCT unit 23 to restore the compressed coded image.

In this case, the motion compensator 24 outputs a frame type memory 25A or 25B or a frame memory in order to immediately output the picture type if the picture type of the reconstructed picture is an I picture or to use the next picture for decoding. 25C).

When the picture type of the currently decoded picture is a B picture or a P picture, both the picture of the picture is reconstructed by using the motion vector included in the bit string and the picture previously decoded and stored.

In general, video compression encoding is intended to improve the transmission rate by obtaining a high compression ratio without damaging the image quality of the original picture. Redundancy should be reduced.

Therefore, in order to reduce redundancy in time and space, as described above, motion is estimated for each frame, a reconstructed image is generated for the estimated motion, and motion vectors are detected and compared with the original image. bar,

As described above, in compression encoding and decoding of video, image compression and reconstruction are performed. The present invention relates to a method for obtaining a motion vector value in order to predict the motion. In general, MPEG2-Video (ISO / Since the motion prediction method in IEC 13818-2 uses block matching, the block matching method also has a different configuration of blocks to perform matching depending on whether the picture type is a frame picture or a field picture. Also in the method, the details vary slightly depending on the method of frame_based, field_based, and dual_prime.

Herein, frame_based motion prediction in a frame picture structure will be described.

3 is a diagram illustrating a method of obtaining a block matching and a motion vector value used for motion prediction. In the related art, a motion vector is obtained in units of 1/2 pixel.

A current frame CF is an image that directly performs image compression, and a reference frame RF is used as a reference to calculate a motion vector MV of the current frame CF. As the image is used, a motion vector and a compression process are sequentially performed on each macro block in the current image frame CF. In this case, block matching is performed on the macro block CM of the current image frame CF. Find the best part.

When the BM drawn in the reference picture frame RF shown in FIG. 3 is the best block matching part, the motion vector of the macro block CM of the current picture frame CF is shown in FIG. Becomes

In this case, for block matching, methods such as full_search, n-step search, telescopic search, etc. are used for a predetermined search window.

When the block matching process is performed as described above, a motion vector is obtained, and the unit of the motion vector is 1 pixel.

For more efficient motion prediction, a motion vector is obtained in 1/2 pixel units. A method of obtaining a motion vector in half pixel units is as follows.

In FIG. 4, the CM virtually represents the current macroblock in the reference image frame RF, and the BM is a position where block matching is performed in units of 1 pixel on the reference image frame RF to obtain the best result.

Here, the size of the macroblock is 16 * 16 pixels, but the description will be made of 8 * 8 pixels.

The position of e becomes a motion vector obtained by 1 pixel resolution, and the positions of a, b, c, d, f, g, h, i become candidates for the motion vector to 1/2 pixel analysis.

Subsequently, block matching is performed again for each of the candidates, wherein the value of the virtual macroblock on the reference frame (RF) used is based on the position of the macroblock and the position of e, respectively. It is obtained by performing interpolation with macroblocks obtained by moving one pixel in the direction of candidates.

In this case, a method of performing interpolation may be performed by equally averaging the values in the two macroblocks.

Through a series of processes described above, a motion vector in units of 1/2 pixels is obtained, and image compression is performed to conform to the syntax defined in MPEG-2.

As described above, in the conventional motion vector detection method, a motion vector in units of one pixel is obtained, and then the interpolation between the macroblock and the macroblock corresponding to the adjacent pixel, that is, the half-pixel using the simple average value between the two macroblocks, is performed. Find a motion vector that is as close as half-pixel.

The present invention improves image compression efficiency by detecting motion vectors in quarter-pixel units using weights according to positions rather than simple average values between macroblocks, and performing image compression accordingly. When video decoding is performed through a decoder, a clearer picture quality can be provided.

The present invention enables to perform image compression by detecting a motion vector up to a quarter pixel unit. The execution structure will be described as follows.

A motion vector is detected for a current video frame from which a current frame to be compressed is predicted from a reference frame, and a motion vector is compressed according to the detected motion vector. In the video compression method used,

The motion vector is configured to perform image compression by detecting a motion vector obtained in units of 1 pixel in units of 1/4 pixels in consideration of weights according to positions with blocks corresponding to adjacent pixels.

The method of detecting a motion vector in units of 1/4 pixels includes: detecting a motion vector in units of 1 pixel;

Obtaining a virtual block on a reference frame by applying weights according to positions of a macroblock corresponding to a detected motion pixel unit and a macroblock corresponding to a pixel adjacent thereto;

Performing block matching between the virtual block on the reference frame obtained in the above process and the macroblock of the current frame;

A motion vector detection process of 1/4 pixel unit is selected to select the best result by comparing the block matching result of the above process with the motion vector of 1 pixel unit obtained in the above process.

FIG. 5 is a diagram for describing a process of calculating a motion vector in units of 1/4 pixels by displaying a macro block as 8 * 8 pixels as in FIG. 4.

First, as in the prior art, block matching is performed on a current video frame to obtain a motion vector in units of 1 pixel. The position b of FIG. 5 is a center position when the motion vector is obtained in units of 1 pixel. .

Then, a method of obtaining a motion vector of 1/2 pixel unit, which has been briefly described in the prior art, will be described first before describing a process of obtaining a motion vector of 1/4 pixel unit.

As described above, motion prediction in units of 1/2 pixel is performed by using a motion vector in units of 1 pixel. An imaginary pixel position at a distance of 1/2 pixel in 8 directions with respect to the position of b is obtained. In this case, 1 ', 2', 3 ', 4', 5 ', 6', 7 ', 8' are determined in eight directions.

When block matching is performed on each of the eight 1/2 pixel vectors determined as described above, the best matching candidate is selected, and the value becomes a motion vector in units of 1/2 pixels.

For example, in the drawing, since the position of 1 'is an intermediate position between the pixels a and b, each value for the block A based on a and the block b based on b in the reference frame The average block obtained on the averaged virtual frame is obtained, and the degree of block matching between the virtual block on the standard frame thus obtained and the macroblock of the current frame is calculated.

Block matching is performed on each position (2 ', 3', 4 ', 5', 6 ', 7', 8 ') in the same manner as described above. By comparing the results at the position of, the best result is the motion vector in half-pixel increments.

In contrast, the method of obtaining a motion vector in units of 1/4 pixels according to the present invention,

If we predict the motion around the position of b in the motion vector of 1 pixel unit and set the position of the virtual pixel at a distance of 1/4 pixel interval in 8 directions, it is 1,2,3,4,5 in 8 each direction. The positions of, 6,7,8 are determined.

By performing block matching on each of the obtained virtual pixels in the same manner as described above, the best matching candidate is selected.

For example, block matching for a position of 1 is located at a distance of 3/4 from a and a quarter from b assuming that the position of 1 is a distance between pixels a and b is 1 Therefore, in the reference frame, the weight is 1/4 for each value of block A based on a, and the weight is 3/4 for each value for block B based on b. The block on an imaginary reference frame calculated by is obtained.

Then, the degree of block matching between the virtual block on the reference frame thus obtained and the macroblock of the current frame is calculated.

Block matching is performed for each position (2,3,4,5,6,7,8) in the same manner as above to compare the results of these eight block matches with the position at b in 1 pixel units. The best result is a motion vector in quarter-pixel increments.

6 is a diagram showing an embodiment of obtaining a motion vector of a unit of 1/4 pixel and another method according to the present invention.

In the embodiment shown in FIG. 6, another 1,2,3,4, at a distance of 1/4 pixel interval in 8 directions by predicting the movement around the position of b in the motion vector in units of 1 pixel The positions of 5, 6, 7, and 8 are defined as virtual pixel positions.

For each virtual pixel thus obtained, the best matching candidate is selected in the same manner as described above. As for the block matching for each virtual position, the distance between pixels a and b is assumed to be 1 as described above. If you do,

In the present embodiment, the positions of the respective virtual pixels are located at a distance of 1/4 from a and at a distance of 3/4 from b, in contrast to the above, and thus block A based on a in the reference frame. For each value of, the weight is 3/4, and for block B based on b, the block on the hypothetical reference frame calculated by calculating the weight is 1/4 for each value is obtained. The same block matching as in the embodiment of FIG. 5 is performed to obtain the best result, thereby obtaining a motion vector in units of 1/4 pixels.

As shown in FIGS. 5 and 6, a motion vector in units of 1/4 pixels can be obtained, and motion vectors in units of 1/4 pixels can be obtained in two ways depending on the encoding capability. The motion vectors can also be obtained by selecting the most accurate values.

When the motion-predicted video frame is compressed and transmitted in the same manner as described above, the decoder can receive an improved image quality compared to the case where a motion vector of 1/2 pixel unit is applied.

In decoding the decoder, compensation is performed for the motion, and the compensation is performed in the same manner as described above.

The motion compensation process will now be described with reference to FIG. 7.

When the exact position of the motion vector of the 1/4 unit received from the decoder is 1, the motion compensation for the motion vector of the 1/4 pixel unit is performed as follows.

A block B is obtained by finding a motion vector of one pixel unit in a reference frame.

A criterion necessary for motion compensation for position 1 of the motion vector in units of 1/4 pixels by applying weights to the block B centered on the position of b obtained above and the block A centered on the position of adjacent a The reference block is calculated.

That is, block A is applied with 1/4 weights for respective values, block B is applied with 3/4 weights (multiplied), and thus the weights of blocks A and B multiplied in this manner are applied. A virtual macroblock is obtained by averaging each value, and motion compensation is performed with the virtual macroblock.

As described above, in the decoder which decodes an image compressed and encoded image using a motion vector of 1/4 pixel unit, the image transmitted from the 1/2 motion vector syntax is converted into a quarter motion vector and processed.

In this case, 1/2 motion vector is 1/2 motion compensation method and 1/4 motion vector is motion compensation method by 1/4 motion compensation method by using the information contained in the bitstream transmitted from the transmitter.

In addition, in this manner, the image compression and reconstruction by applying the motion vector in the unit of 1/4 pixel is made. As described above, the 1/2 motion vector is applied without supporting both 1/2 and 1/4. In order to maintain compatibility with the existing decoder that performs the motion compensation, the encoder transmits the same syntax as the 1/2 motion vector, and the existing decoder recognizes the motion as 1/2 motion vector and performs the motion compensation as before. By doing so, it is possible to maintain compatibility with the existing standard ISO / IEC 13818-2 (MPEG-2).

That is, 1 ', 2', 3 ', 4 predicting the movement in 1/2 pixel units instead of the positions of 1,2,3,4,5,6,7,8, b as shown in FIG. The motion vectors corresponding to 1/2 pixel units corresponding to ', 5', 6 '., 7', b are transmitted correspondingly.

At this time, if the motion vector is predicted by 1/4 and the motion compensation is performed by the 1/4 motion compensation method, a good result is obtained. Compensation is performed, but error propagation may occur when the GOP (Group of picture) of the image increases.

In order to prevent such a case, an image compression algorithm may be improved in consideration of accumulation of errors in image compression.

As described above, the interpolation between the macroblock for the motion vector in units of 1 pixel and the macroblock corresponding to the adjacent pixel is not a simple average value, but the weight of the motion vector is reduced by at least 1/4 pixel. By calculating up to a unit, the image compression efficiency is increased, so that the improved image quality can be provided after the image decoding is completed.

Claims (3)

A decoding method of a video compressed in block units; In order to obtain a reference block for reconstructing the current block of the image and to perform motion compensation in units of 1/4 pixels interpolated in units of 1 pixel, 1/4 based on a motion vector of 1/4 pixels and the reference block Restoring the current block by applying pixel-by-pixel motion compensation; The current block recovery step, Applying preset weights to pixel values at pixel positions corresponding to one pixel units in the reference block; Averaging the weighted pixel values to obtain pixel values of pixel positions corresponding to quarter pixel units within the reference block; and Restoring a current block by using a virtual block including a pixel value of a pixel position corresponding to the quarter pixel unit, And the predetermined weight is determined according to a pixel position corresponding to one pixel unit in the reference block. The method of claim 1, wherein the obtaining of the pixel value at the pixel position corresponding to the quarter pixel unit comprises: And a pixel value to which the weight is applied and a reference pixel value of the reference block are averaged. The method of claim 1, The method of claim 1, wherein the motion vector in units of 1/4 pixels is derived from a motion vector in units of 1 pixel.

Images