KR0134700B1 - Method for estimating moving pixels - Google Patents

Abstract

The present invention provides an efficient half-pixel motion estimation apparatus for an image compression method, comprising: address generation means (3-1) for outputting a memory address; A storage means (3-2) used to calculate a motion compensation error at each half-pixel position by receiving the memory address output from the address generating means (3-1) and outputting an integer position motion compensation error; A calculator (3-3) which receives motion compensation errors at integer locations output from the memory and calculates motion compensation errors at eight half-pixel locations around the integer location having the minimum motion compensation error; And a comparator for comparing the eight half-pixel position motion compensation errors outputted from the calculator 3-3 and the minimum motion compensation errors at integer positions to output the position having the minimum motion compensation error as a final half-pixel unit motion vector. And a half-pixel unit motion vector selector (3-4).

Description

An Efficient Half-Pixel Motion Estimator in Image Compression Method

1 is a flowchart of a motion estimation unit according to the present invention;

2 is a modeling diagram for half-pixel unit motion compensation error,

3 is a block diagram of a motion estimator according to the present invention;

Explanation of symbols on the main parts of the drawings

3-1: Address generating means

3-2: Memory

3-3: Operator

3-4: Comparator and Half-Pixel Motion Vector Selector

The present invention relates to a half-pixel motion that effectively reduces a large amount of computation required for half-pixel motion estimation and provides improved performance over the conventional method.

In order to estimate half-pixel motion, an error must be obtained at each half-pixel position around an integer motion vector. At this time, since the error of each half-pixel position is obtained by the full search method, it requires a large amount of computation. Obtain

The first commonly used method for motion estimation in half-pixel units is to obtain a motion vector in integer pixel units, obtain the values of half-pixel positions around the vector by interpolation, and then use the periphery including the integer-pixel motion vectors obtained earlier. The second full search is performed on the half-pixel unit vectors, and the position with the smallest error is determined as the half-pixel unit vector.

However, the half-pixel motion estimation method using on-a-field searching requires a large amount of hardware as well as additional time in implementation.

The second method approximates the error at the half-pixel position by using the model of the motion compensation error obtained at the process of obtaining the motion vector in the integer pixel unit and the motion compensation error at the half-pixel position, and uses the errors to use the half-pixel. A method of performing motion estimation.

In this method, the performance depends on the model of the error. In the conventional method, it is difficult to expect any more performance by assuming that the motion compensation error converges in all directions about the minimum error when setting the model.

In order to solve the above problems, the present invention provides a half-pixel motion estimation method and method for reducing the amount of hardware and computation time and further improving performance through modeling considering the degree of convergence of errors in each direction. There is a purpose.

In order to achieve the above object, the half-pixel motion estimation apparatus of the present invention comprises: address generation means for outputting a memory address; Storage means for receiving a memory address output from the address generating means and outputting an integer position motion compensation error used to calculate a motion compensation error at each half-pixel position; A calculator for receiving motion compensation errors at integer positions output from the memory and calculating motion compensation errors at eight half pixel positions around the integer position having the minimum motion compensation error; And a comparator and a half-pixel unit motion outputting the position having the minimum motion compensation error as a final half-pixel unit motion vector by comparing the eight half-pixel position motion compensation errors output from the calculator and the minimum motion compensation errors at integer positions. It comprises a comparator and a half-pixel unit of motion vector selector (3-4) for outputting as a vector.

In addition, the half-pixel motion estimation method of the present invention includes a first step of obtaining an integer unit motion vector by the motion estimation method; A second step of determining whether the motion estimation method is a complete search method; And if the determination result is a full search method, the half-pixel unit motion estimation is performed using the motion vector obtained in the first step, and if the determination result is not a full search method, at a constant position around the motion vector obtained in the first step. And a third step of performing a half-pixel unit motion estimation by obtaining an error of.

Hereinafter, the present invention will be described in detail with reference to FIGS. 1 through 3.

First, the motion estimation method for the present invention is configured as the first diagram.

As shown in the figure, an integer unit motion vector is obtained by various motion estimation methods such as full search or hierarchical search (1-1). At this time, the search area becomes H1 in the horizontal direction and V1 in the vertical direction. When the method used for the motion estimation is the full search method, the motion compensation error is obtained at all integer positions in the search area. However, when hierarchical search or other motion estimation method is used, the integer position error around the motion vector to be used for half-pixel motion estimation is not obtained.

Therefore, in this case, the error at the constant position around the motion vector obtained in (1-1) is further obtained from (1-2). At this time, the search area becomes H2 in the horizontal direction and V2 in the vertical direction. In (1-3), the motion estimation in half pixel units is performed by using the obtained motion vector and the error in the surrounding integer position.

A method of performing half-pixel unit motion estimation of FIG. 1 will be described with reference to FIG. 2 as follows.

The circle in the center and the surrounding 16 circles shown in FIG. 2 (a) are the minimum motion compensation error positions obtained by performing (1-1) and (1-2) in FIG. Indicates error location.

The eight squares around the center of the circle represent half-pixel positions.

Errors at each half-pixel position are obtained as approximations by the model of the present invention as follows.

Diagonal P or X shown in FIG. 2 (a) represents the magnitude of the motion compensation error at each position.

FIG. 2 (b) shows four integer position errors P (-2), P (-1), P (0) and P (1) which are used to calculate the magnitude X (-0.5) of the half-pixel motion compensation error. Are redrawn with the location and the magnitude of the error as the axis.

The half-pixel motion compensation error X (-0.5) to be obtained by referring to FIG. 2 (b) is modeled as follows.

X (-0.5) = a (A + B) + C (1)

In Equation (1), A and B become linear predicted values according to the convergence degree of the motion compensation error in each direction as shown in FIG. In other words,

A = P (-1) -abs (P (-2) -P (-1) / 2

B = P (0)-(P (1) -P (0) / 2 (2)

a is the weighting factor determined experimentally. C is an offset for correcting the predicted value at each half pixel position, and is determined by four integer motion compensation errors, both of which are based on the currently predicted half pixel position.

The models of equations (1) and (2) apply equally to the diagonal, horizontal and vertical directions, and approximate the motion compensation errors at all eight half-pixel positions. The position having the minimum error among the obtained errors of the eight half-pixel positions and the exact center position becomes the motion vector of the final half-pixel unit.

As an example of applying the models of equations (1) and (2), let a = 1/4 and set the value of C

C = (P (-1) + P (0)) / 4 (3)

In this case, the half-pixel motion compensation error X (-0.5) is obtained by the following equation (4).

X (-0.5) = P (-1) / 2 + 5P (0) / 8-P (1) / 8-abs (P (-2) -P (-1) / 8 (4)

In the same way, half-pixel motion compensation error X (

X (+0.5) = P (1) / 2 + 5P (0) / 8-P (-1) / 8-abs (P (-2) -P (-1) / 8

The half-pixel motion estimation method based on the models of Equations (1) and (2) provides satisfactory performance with only a small amount of computation and has an advantage that it can be easily implemented with a small amount of hardware.

Next, the half-pixel unit motion estimation of the half-pixel unit motion estimation apparatus according to the present invention will be described with reference to FIG. 3.

First, the address generator 3-1 generates a memory address for outputting integer position motion compensation errors from the memory 3-2, which are used to calculate the motion compensation error at each half pixel position.

The calculator (3-3) receives the motion compensation errors at the integer positions output from the memory and uses the models of Equations (1) and (2) to calculate eight halves around the integer positions having the minimum motion compensation errors. The motion compensation error at the pixel position is calculated and output to a comparator and a half pixel unit of motion vector selector 3-4.

Finally, the comparator and the half-pixel unit motion vector selector 3-4 compare the eight half-pixel position motion compensation errors output from the calculator 3-3 and the minimum motion compensation errors at the integer positions with each other. The position with the error is output as the final half-pixel unit motion vector.

The half-pixel motion estimation apparatus and method thereof according to the present invention have the effect of reducing the amount of hardware and the computation time and further improving performance through modeling considering the degree of convergence of errors in each direction.

Claims (4)

In an efficient half-pixel motion estimation apparatus for an image compression method, Address generation means 3-1 for outputting a memory address; A storage means (3-2) which receives the memory address output from the address generating means (3-1) and outputs an integer position motion compensation error used to calculate a motion compensation error at each half pixel position; A calculator (3-3) which receives motion compensation errors at integer locations output from the memory and calculates motion compensation errors at eight half-pixel locations around the integer location having the minimum motion compensation error; And A comparator for comparing the eight half-pixel position motion compensation errors output from the calculator 3-3 with the minimum motion compensation errors at integer positions and outputting the position having the minimum motion compensation error as a final half-pixel unit motion vector; Half-Pixel Motion Vector Selector (3-4) Half-pixel motion estimation device comprising a. In the efficient half-pixel motion estimation method in the image compression method, A first step (1-1) of obtaining a motion vector by an integer unit by a motion estimation method; A second step of determining whether the motion estimation method is a complete search method; And the half-pixel motion estimation is performed using the motion vector obtained in the first step if the judgment result is a complete search method, and at a constant position around the motion vector obtained in the first step if the search result is not a full search method. The third step of performing the half-pixel unit motion estimation (1-3) by obtaining the error of (1-2) Half-pixel motion estimation method comprising a. The method of claim 2, wherein the search region of the first step is a first horizontal direction and a first vertical direction. 3. The half-pixel motion estimation method of claim 2, wherein the search area in the third step that is not the full search method is the second horizontal direction and the second vertical direction.