KR0185841B1 - A motion extimator for a video coder - Google Patents

Abstract

The present invention relates to a motion estimation apparatus in an image encoder. The apparatus of the present invention is a motion estimation apparatus in an image encoder including an integer pixel motion estimation memory 10 and an integer pixel motion estimation unit 12. A half-pixel motion estimation memory 20 for storing image data corresponding to a search region from the integer pixel motion estimation memory 10 for motion estimation in units of half pixels; And a half-pixel motion estimator 22 for performing half-pixel-based motion estimation around an integer-based motion vector and performing half-pixel-based motion estimation around the motion vector, and outputting a motion vector. There is an advantage that the cost of hardware can be reduced by using the memory of the search area capacity instead of the memory of the frame capacity as the memory for half-pixel motion estimation.

Description

Motion Estimator in Image Encoder

1 is an exemplary diagram illustrating motion estimation in units of half pixels.

2 is a block diagram of a motion estimation apparatus in a conventional image encoder.

3 is an exemplary diagram illustrating integer pixels required for one macroblock processing in a half-pixel motion estimation unit.

4 is a block diagram of a motion estimation apparatus in an image encoder of the present invention.

5 is an exemplary diagram illustrating a search region in half pixel motion estimation.

* Explanation of symbols for main parts of the drawings

10: memory for integer pixel motion estimation

12: integer pixel motion estimation unit

20: memory for half pixel motion estimation

22: half-pixel motion estimation unit

The present invention relates to a motion estimation apparatus, and more particularly, to a motion estimation apparatus in an image encoder configured to perform motion estimation with half-pixel accuracy.

Video encoders that use the Moving Picture Experts Group-2 (MPEG-2) method, such as HDTV (High Definition TV) or video phones, undergo various data compression processes to reduce the data to be transmitted to the video encoder. One method of the data compression process is motion estimation.

In general, motion estimation is an estimation of a motion vector indicating how much the pixels of the current frame move with respect to the pixels of the previous frame in a continuous image signal, and transmits the entire image data. Instead, it refers to an image data compression technique that reduces the amount of image data transmitted by transmitting such motion vectors.

The motion estimation method in each method is as follows.

1) ATRC method (ADTV; Advanced TV)

As for the motion estimation method used in ADTV, the algorithm proposed in MPEG adopts a block matching algorithm (BMA) with half-pixel accuracy.

The Block Matching Algorithm (BMA) assumes that the motion of the screen is moved horizontally or vertically in parallel, so that the block image of the frame in which the movement occurred (i.e., the current frame) is at a position in the frame (i.e., the previous frame) before the movement occurs. It is a method of estimating a motion vector through its position by estimating whether it best matches a block picture.

Motion estimation in ADTV refers to the process of calculating a two-dimensional vector representing the degree of motion, which is called a motion vector.

Two types of motion estimation are used in ADTV, one for forward motion estimation using the past screen to make the current screen, and the other for backward motion estimation using the future screen to make the current screen. To do.

Both methods find the motion vector by finding the block with the least distortion in the predetermined search interval with M * N.

Popular error (or distortion) measurement methods include Mean Squre Error (MSE) and Mean Absolute Error (MAE). ) Is mainly used.

Half-pixel motion estimation consists of two steps: the first search is to obtain an integer-pixel motion vector and the half-pixel motion vector is centered around the first pixel-by-pixel motion vector obtained by the second search. Will be obtained.

In ADTV, input video is divided into several types and divided into frames. There are three types of I-frames, P-frames, and B-frames. Looking at it as follows.

I-frame does not estimate the motion, so there is no motion vector, but it is used to estimate the motion of another frame. P-frame uses the previous I-frame or P-frame to estimate forward motion, and another P-frame or B-frame. Used for motion estimation, B-frames use forward and backward I-frames or P-frames to perform forward motion estimation and backward motion estimation, and interpolate by dividing the forward motion compensated frame and the backward motion compensated frame in two. The motion vector is determined by selecting the smallest difference between the displaced block differences (DBD) among the three motion estimation results.

When the smallest value of the DBD is interpolated, both a forward motion vector and a backward motion vector are transmitted.

FIG. 1 is an exemplary diagram showing motion estimation in units of half pixels. Referring to FIG. 1, half-pixel motion estimation will be described below.

After a full search for integer-pixels, a block matching algorithm (BMA) is performed on eight half-pixel positions.

The spatially interpolated pixel value is calculated as follows.

2) G.I Company (Digicipher)

The motion estimation method shown in G.I's Digicipher system proposal is similar to the block matching algorithm (BMA) used in ATRC's ADTV described above, but is less than ADTV because it does not estimate half-pixel motion in accuracy.

In other respects, there is no B-frame of ADTV by performing forward motion estimation without backward motion estimation. That is, Digicipher's frame consists of only I-frames and P-frames.

The unit of motion estimation is a macro block, which is composed of 4 * 2 Y (luminance signal) blocks, 1 U (color difference signal) block, and 1 V (color difference signal) block. Have.

3) MIT method (ATVA-Progressive HDTv)

The motion estimation method proposed by MIT is based on a spatial-temporal constraint, which assumes that image displacement occurs at a constant rate between frames, and correlates the displacement rate with the motion vector. It is a method to estimate the motion vector by finding a spatial-temporal constraint, which is a differential equation that represents.

This method assumes a uniform speed of transmission and is not valid for object rotation, camera zoom, moving object transmission areas, or multi-object motion with different motion vectors.

Even if the motion is assumed to be uniform within a small interval and the motion estimation is suppressed in the range where the motion estimation is uncertain, the correlation is greatly reduced by time due to the motion estimation and compensation based on the spatial-temporal constraint. do.

Compared to the Block Matching Algorithm (BMA), this method greatly reduces the amount of computation depending on the block size and performs well with and without noise.

The motion vector is estimated from the luminance signal and the same motion vector is used for the luminance signal Y and the chrominance signal U and V, with one dominant vector obtained for every 32 * 32 pixel block.

Prediction is performed on the current frame from the previous coded frame and the motion vector, and a difference between the current frame and the predicted code is calculated. This difference is called a motion compensated residual.

This is obtained for each of the three parts of Y. U. V and motion compensation is invalidated and the picture frame itself is encoded when it has relatively large energy for the picture frame (when there is a picture change).

4) Zenith and ATT (DSC-HDTV)

Due to the nature of the image, the pixel values of adjacent pixels are generally composed of very close values because the pixels in the line representing the boundary and contour of the object are used to display the same object.

Zenith and ATT used a hierarchical mothion estimation method that effectively estimates motion by eliminating unnecessary redundancy of frame image data by using the characteristics of these images.

In addition, in order to increase the efficiency of the transmission channel, Zenith and ATT have proposed two-stage motion estimation of coarse motion estimation and fine motion estimation, which are described in detail as follows. .

First, the filter is used to remove redundancy between the frame video signal and the previous video signal. As a result, the resolution is reduced by 1/2 each in the horizontal and vertical directions.

Coarse motion estimation from low-resolution video signals with a total area reduced to 1/4 estimates an absolute value of the pixel value difference of the previous frame from the current frame in units of 32 * 16 (horizontal * vertical) pixels, almost in blocks. This is done by comparing the errors, using only the components of the frequency and comparing the estimation error to the previous frame block, where the smallest of the estimated errors obtained by using only the luminance component and moving the previous frame block by the search interval Becomes the motion vector of.

The block matching in the hierarchical low resolution image as described above reduces the search interval and the block size by a quarter compared to that used in the original high resolution image, thereby significantly reducing the calculation amount.

High resolution motion estimation uses motion vectors obtained from low resolution images to replace them with motion vectors in the original high resolution images and subdivides them into high resolution motion vectors.

That is, since the 16 * 8 block size and the low resolution motion vector in the low resolution image are each reduced by 1/2 in the horizontal and vertical directions of the image having the original resolution, double the both to obtain the block size and the motion vector of the original image.

It is divided into 8 * 8 small block sizes, and each small block is moved back from the low resolution motion vector by the search interval to obtain a high resolution motion vector.

The motion vector encoder estimates the screen of the next frame using motion vectors estimated at low resolution and high resolution, respectively, and selects a motion vector corresponding thereto.

This is for limiting the bit rate in compressing the motion vector, and depending on the margin of the channel band width, it may send only a low resolution motion vector or a low resolution and high resolution motion vector.

There are several methods for motion estimation as described above. In MPEG-2, motion estimation is performed with half-pixel accuracy. This function uses the resultant motion vector after motion estimation is completed in integer pixel units. Thus, half-pixel unit motion estimation is performed.

2 is a block diagram of a motion estimation apparatus in a conventional image encoder, and corresponds to an I-picture and a P-picture of image data input for motion estimation in units of integer pixels. An integer pixel motion estimation memory 10 for storing data of an anchor frame; An integer pixel motion estimation unit (12) which receives data of a current frame and search area data from the integer pixel motion estimation memory (10), performs motion estimation in units of integer pixels, and outputs a motion vector; A half-pixel motion estimation memory 14 for storing data of an anchor frame corresponding to an I-picture and a P-picture among image data inputted for half-pixel motion estimation; And a half-pixel motion estimation unit 16 for performing motion estimation in half-pixel units based on the motion vector in integer units and outputting the motion vector.

FIG. 3 is a diagram illustrating integer pixels required for one macroblock processing by a half-pixel motion estimation unit. Referring to FIG. 1 and FIG. 2, operation of a motion estimation apparatus in a conventional image encoder is as follows.

Among the input image data, data of an integer frame corresponding to an I-picture and a P-picture are stored in the integer pixel motion estimation memory 10 and the half-pixel motion estimation memory. Each of the memory units 10 and 14 stores image data in units of frames.

The integer pixel motion estimation unit 12 receives the data of the current frame as a reference, receives data of the search area from the integer pixel motion estimation memory 10, performs motion estimation in units of integer pixels, and then most similarly. Find the data and output the motion vector for it.

The motion vector of the integer unit obtained as described above is transferred to the half-pixel motion estimation unit 16, and the half-pixel motion estimation unit 16 includes nine half-pixel units as shown in FIG. Motion estimation is performed for the case of.

In order to estimate a half-pixel motion of 1 macroblock (16 pixels * 16 pixels), two more pixels are required, horizontally and vertically, respectively, as shown in FIG. 3. In the nine cases as shown in FIG. 1, the difference between the current frame is compared and the most similar one, that is, the one having the smallest error value is selected, and this is added to the least significant bit (LSB) of the motion vectors in integer units and output.

As described above, when a motion estimating device that performs motion estimation in units of half pixels is composed of real hardware, a problem arises that it becomes very complicated and bulky. In particular, a conventional motion estimating device uses a frame unit for half pixel motion estimation. There is a problem in that the cost of hardware is increased because of using a memory having a capacity.

Accordingly, an object of the present invention is to provide a motion estimation apparatus in an image encoder, which is designed to solve the above problems, and to receive and store image data for half-pixel motion estimation in units of search areas. There is this.

The motion estimation apparatus according to the present invention for achieving the above object, an integer pixel for storing the data of the reference frame corresponding to the I-picture and P-picture of the image data input for motion estimation in the unit of integer pixels In an image encoder including a motion estimation memory, data of a current frame, and search region data from the integer pixel motion estimation memory, and performing an integer pixel motion estimation and outputting a motion vector. In the motion estimation apparatus of

A half-pixel motion estimation memory for storing image data corresponding to a search region from the integer pixel motion estimation memory for motion estimation in half-pixel units; And

And a half-pixel motion estimator for outputting a motion vector after performing motion estimation in half-pixel units around a motion vector in an integer unit.

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

4 is a block diagram of a motion estimation apparatus in an image encoder according to the present invention. The motion estimation apparatus of the present invention corresponds to an I-picture and a P-picture of image data input for motion estimation in units of integer pixels. An integer pixel motion estimation memory 10 for storing data of a reference frame; An integer pixel motion estimation unit (12) which receives data of a current frame and search area data from the integer pixel motion estimation memory (10), performs motion estimation in units of integer pixels, and outputs a motion vector; A half-pixel motion estimation memory 20 for receiving and storing image data corresponding to a search region from the integer pixel motion estimation memory 10 for motion estimation of a half pixel unit; And a half-pixel motion estimation unit 22 that performs motion estimation in half-pixel units based on the motion vector in integer units and outputs the motion vector.

Next, the operation and effects of the present invention will be described in detail with reference to the accompanying drawings.

FIG. 5 is an exemplary diagram illustrating a search area when the half-pixel motion is estimated. Referring to FIG. 1, FIG. 5 will be described with reference to FIGS. 1, 3, and 4.

Among the input image data, data of an anchor frame corresponding to an I-picture and a P-picture are stored in the integer pixel motion estimation memory 10, which stores image data in units of frames. do.

The integer pixel motion estimation unit 12 receives the data of the current frame as a reference, receives data of the search area from the integer pixel motion estimation memory 10, performs motion estimation in units of integer pixels, and then closes the most similarly. Find the data and output the motion vector for it.

The half-pixel motion estimation memory 20 receives and stores image data in a search area unit from the integer pixel motion estimation memory 10, and stores the image data in the integer vector motion vector output from the integer pixel motion estimation unit 12. Accordingly, the half pixel motion estimation unit 22 outputs necessary data as shown in FIG. 3.

As described above, instead of using the frame capacity memory as the half-pixel motion estimation memory 20, the cost of hardware can be reduced.

However, as shown in FIG. 5, if the motion vector outputted after performing the integer pixel motion estimation is at the edge of the search region of the integer pixel motion estimation, external pixels of the search region are also necessary for half pixel motion estimation. It happens.

However, in the half-pixel motion estimation memory 20 that stores only the data in the search area, since the pixels outside the search area do not exist as described above, when the motion vector after performing the integer pixel motion estimation is at the edge, it is shown in FIG. Of the nine cases, if the pixel outside the search area is required, the half-pixel motion estimation is performed by considering only the case in which the pixel inside the remaining search area can be calculated only.

Even if the half-pixel motion estimation is performed in this way, the motion vector in the integer pixel motion estimation is rarely present at the edge of the search area. Even if the motion vector is generated, in some of the nine cases of FIG. It does not have a big impact on image quality because it is not forcibly generated.

Eventually, there is some loss, but the use of memory is greatly reduced, which can reduce the cost of hardware.

As described above, the present invention has the effect that the cost of hardware can be reduced by using the memory of the search area capacity instead of the memory of the frame capacity as the memory for half-pixel motion estimation.

Claims (1)

Integer pixel motion estimation memory 10 for storing data of reference frames corresponding to I-pictures and P-pictures among image data input for motion estimation in units of integer pixels, data of a current frame, and the integer pixel motion estimation A motion estimation apparatus in an image encoder including an integer pixel motion estimation unit 12 that receives search area data from the memory 10 and performs motion estimation in units of integer pixels, and outputs a motion vector. A half-pixel motion estimation memory 20 for storing image data corresponding to a search area from the integer pixel motion estimation memory 10 for motion estimation of a unit; And a half-pixel motion estimation unit (22) configured to output a motion vector after performing motion estimation on a half-pixel basis around a motion vector in an integer unit.