KR101368732B1 - Apparatus for estimating motion for h.264/avc encoder with high performance and method thereof - Google Patents

Abstract

The present invention relates to an apparatus for estimating motions for H.264/AVC encoder with high performance which can minimize damages to the images by detecting median and directional nature of movement vectors of adjacent macroblocks when estimating motions, estimating initial searching points, and reducing computational complexity; and can reduce the memory size when performing hardware, and to a method for estimating motions for H.264/AVC encoder. The apparatus for estimating motions for H.264/AVC encoder with high performance comprises: a motion estimation control unit which computates mvp from motion vectors of three adjacent macroblocks for current macroblock in order to set up an initial searching point; an integer pel motion estimation unit which produces the integer pel motion vectors for current macroblock using the initial searching point; and a fractional pel motion estimation unit which produces fractional motion vectors for current macroblock using pixels of reference frame corresponding to the integer motion vector produced from the integer motion estimation unit. [Reference numerals] (100) Motion estimation control unit; (200) Integer pel motion estimation unit; (300) Fractional pel motion estimation unit; (400) Motion vector ram

Description

Motion Prediction Device for High Performance H.264 / ACC Encoder and Motion Prediction Method {APPARATUS FOR ESTIMATING MOTION FOR H.264 / AVC ENCODER WITH HIGH PERFORMANCE AND METHOD THEREOF}

The present invention relates to a motion prediction apparatus and a motion prediction method for an H.264 / AVC encoder. More specifically, the present invention predicts an initial search point by detecting an intermediate value and a direction of motion vectors of a neighboring macroblock during motion prediction. In addition, the present invention relates to a motion prediction apparatus and a motion prediction method for a high performance H.264 / AVC encoder that can reduce image damage by reducing the amount of computation associated with motion prediction and reduce the memory size in hardware implementation.

With the recent development of broadband transmission, wired or wireless broadband connection lines have been speeded up and storage media have become large in capacity. In addition, multimedia such as text, voice, pictures, and videos also require better color and image quality, and the amount of information used is increasing with the times.

In particular, when the information to be transmitted is a large amount of information generated compared to the audio, there is a lot of difficulty in directly transmitting or storing the image data. Therefore, a situation in which high image compression efficiency is being developed.

H.264 / AVC is an international video compression standard developed jointly by ITU-T and ISO / IEC. It is a digital video codec standard with very high data compression ratio. It is the existing standard MPEG-1, MPEG-2, MPEG- 4 Compared to Part 2, it was developed to get low bit rate at the same quality. This H.264 / AVC provides about twice as much compression efficiency as the MPEG-4 video standard.

In particular, H.264 / AVC introduces new compression methods such as 4x4 block motion compensation, 1/4 pixel motion prediction, improved entropy coding, and deblocking filter to improve the compression efficiency of images.

Motion prediction technology is an important process that generally takes up more than 60% of the computation in image compression. In H.264 / AVC, motion prediction technology can use frame correlation to remove temporal redundancy between frames.

Techniques related to motion prediction to reduce the amount of computation have been actively developed, and the motion vector field adaptive search technique (MVFAST), predictive motion vector field adaptive search technique (PMVFAST), and enhanced predictive zonal search (EPZS) are currently macroblocks. The initial search point is set by using that the motion vector of C is highly correlated with the motion of the neighboring macroblocks.

However, MVFAST can increase the search point by a two-stage diamond search method, and PMVFAST and EPZS use the motion vector of the previous frame, which causes a problem of increasing memory size in hardware implementation.

In addition, in relation to a technology capable of reducing memory capacity, Korean Patent Laid-Open Publication No. 10-2008-0067922 relates to an image decoding method and apparatus having an image reduction function. The decoded image signal is classified into an intra block and an inter block. In addition, a technique for performing a motion adaptive image reduction function by selectively performing anti-aliasing filtering on the divided intra block image and the inter block image is disclosed.

However, the above-described prior art does not include a technique for predicting the initial search point according to the median and the direction of the motion vectors of the principal surface macroblock during motion prediction.

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and is a high performance motion prediction method for H.264 / AVC that can predict an initial search point according to the median and the direction of motion vectors of neighboring macroblocks during motion prediction. And a motion predicting apparatus.

In addition, the present invention provides a high-performance motion prediction method and a motion prediction apparatus for H.264 / AVC that can reduce the amount of computation and minimize image damage by predicting the initial search point more accurately using the motion vectors of the neighboring macroblocks. It is for the purpose of providing.

In addition, an object of the present invention is to provide a high-performance motion prediction method and a motion prediction apparatus for H.264 / AVC capable of reducing the capacity of the entire memory by storing only motion vectors of neighboring macroblocks by performing motion prediction with a small amount of computation. It is done.

However, the object of the present invention is not limited to the above-mentioned objects, and other objects not mentioned can be clearly understood by those skilled in the art from the following description.

In order to achieve the above object, the motion prediction apparatus for the high performance H.264 / AVC encoder according to the present invention calculates an intermediate motion vector value (mvp) from the motion vectors of three neighboring macroblocks for the current macroblock, and performs an initial search. A motion prediction control unit for setting a point, a refiner motion predictor for generating a refiner motion vector for the current macroblock using the initial search point, and a pixel of a reference frame corresponding to the refiner motion vector generated by the refiner motion predictor And a subpixel motion predictor for generating a subpixel motion vector for the current macroblock.

In addition, the motion prediction apparatus according to the present invention, the motion prediction controller, the intermediate motion vector for calculating the intermediate motion vector value (mvp) by using the magnitude and direction of the motion vectors of the three neighboring macroblocks for the current macroblock A value generation module, a search point mode selection module for setting a search point mode and an initial search point according to the direction of the intermediate motion vector value, and data of a current macroblock and data of a reference frame from an external memory and stored in an internal memory It characterized in that it comprises a memory control module.

In addition, the motion prediction apparatus according to the present invention is characterized in that the intermediate motion vector value generation module calculates the intermediate motion vector value using the following relational expression.

Where MV _left is the motion vector of the left macroblock for the current macroblock, MV _top is the motion vector of the upper macroblock for the current macroblock, and MV _{top - right} is the _{top - right} macroblock of the current macroblock. Represents a motion vector)

In addition, the motion prediction apparatus according to the present invention is characterized in that the search point mode selection module sets a search point mode by comparing the x coordinate and the y coordinate of the intermediate motion vector value through the following relational expression.

In addition, the motion prediction apparatus according to the present invention, in the search point mode selection module, if the intermediate motion vector value has a linear direction, the diamond search pattern, and if the intermediate motion vector value has a diagonal direction, the search point mode is selected with a square search pattern. And the search point mode is set in nine modes according to the direction of the intermediate motion vector value.

In addition, the motion prediction apparatus according to the present invention is characterized in that the search point mode selection module sets an initial search point to at least five to nine search points according to the search point mode to be set.

In addition, the motion prediction apparatus according to the present invention, the refinery motion prediction unit, the SAD calculation module for calculating the SAD (Sum of Absolute Difference) value for each initial search point and the calculated SAD value by comparing the minimum SAD value A minimum value selection module for storing each macroblock size and selecting a macroblock having a minimum SAD value, and a refinement motion vector generation module generating a refinement motion vector for the macroblock having the minimum SAD value selected by the minimum value selection module. Characterized in that.

In addition, the motion estimation apparatus according to the present invention is characterized in that the SAD calculation module calculates the SAD value through the following relational expression.

Where F _t represents the current frame and F _{t -i} represents the reference frame

In addition, the motion prediction apparatus according to the present invention, when the size of the macroblock having the minimum SAD value is 16 × 16, the motion vector generating module generates one motion vector, and the macroblock having the minimum SAD value When the size is 4 × 4, 16 motion vectors are generated.

In addition, the motion prediction apparatus according to the present invention includes interpolation in which the subpixel motion predictor receives 16 pixels of a reference frame corresponding to the refiner motion vector and interpolates in quad pixel units to generate 16 × 16 quad pixel data. A SAD calculation module for calculating a SAD value by performing a SAD operation on the generated 16 × 16 quad pixel data for one cycle, and storing the calculated SAD value and having a minimum SAD value among the calculated SAD values. And a subpixel motion vector generation module for generating a subpixel motion vector of the macroblock.

In addition, in the motion prediction method for the high performance H.264 / AVC encoder according to the present invention, the motion prediction controller uses an intermediate motion vector value (mvp) by using the magnitude and the direction of the motion vectors of three neighboring macroblocks with respect to the current macroblock. A first step of calculating a second step; a second step of setting a search point mode and an initial search point from the intermediate motion vector value calculated by the motion prediction controller; and a refinement motion predictor performing an initial search point set in the second step. A third step of generating a refinement motion vector for the current macroblock based on the fourth step and a fourth step of generating a subpixel motion vector for the current macroblock using the refinement motion prediction vector generated in the third step. Characterized in that it comprises a step.

According to the motion prediction apparatus and the motion prediction method for the high-performance H.264 / AVC encoder of the present invention, a large number of search points can be obtained by predicting the initial search point according to the direction and the median value of the motion vectors of the neighboring macroblocks during the motion prediction. There is an advantage to having a search area.

In addition, according to the motion prediction apparatus and the motion prediction method for the high performance H.264 / AVC encoder of the present invention, by using the motion vector of the neighboring macroblocks more accurately predict the initial search point to reduce the amount of computation and to minimize the damage of the image In addition, since the motion vector of the neighboring macroblock is used, the loss of the peak signal to noise ratio (PSNR) is small compared to the amount of computation even in the HD resolution.

1 is a block diagram showing the configuration of a motion prediction apparatus for a high performance H.264 / AVC encoder according to the present invention.
2 is a block diagram showing the configuration of a motion prediction controller according to the present invention.
3 is a block diagram showing the configuration of a refinery motion prediction unit according to the present invention.
4 is a block diagram showing a configuration of a subpixel motion prediction unit according to the present invention.
5 is a flowchart illustrating a motion prediction method for a high performance H.264 / AVC encoder according to the present invention.
6 is a flowchart illustrating a method of generating a refiner motion vector for a current macroblock according to the present invention.
7 is a flowchart illustrating a method of generating a subpixel motion vector for a current macroblock according to the present invention.
8 is an exemplary diagram illustrating a motion vector of a neighboring macroblock with respect to the current macroblock.
9 is an exemplary view illustrating an initial search point mode in the motion prediction controller according to the present invention.
10 is an exemplary diagram illustrating a SAD calculation module for calculating a sum of absolute difference (SAD) in units of 4x4 macroblocks according to the present invention.
11 is an exemplary diagram illustrating a structure of half pixel interpolation according to the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, a detailed description of preferred embodiments of the present invention will be given with reference to the accompanying drawings. In the following description of the present invention, detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear.

Embodiments in accordance with the concepts of the present invention can make various changes and have various forms, so that specific embodiments are illustrated in the drawings and described in detail in this specification or application. It is to be understood, however, that it is not intended to limit the embodiments according to the concepts of the present invention to the particular forms of disclosure, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention.

It is to be understood that when an element is referred to as being "connected" or "connected" to another element, it may be directly connected or connected to the other element, . On the other hand, when an element is referred to as being "directly connected" or "directly connected" to another element, it should be understood that there are no other elements in between. Other expressions that describe the relationship between components, such as "between" and "between" or "neighboring to" and "directly adjacent to" should be interpreted as well.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this specification, the terms "comprises ",or" having ", or the like, specify that there is a stated feature, number, step, operation, , Steps, operations, components, parts, or combinations thereof, as a matter of principle.

1 is a block diagram showing a configuration of a motion prediction apparatus for a high performance H.264 / AVC encoder according to the present invention, FIG. 2 is a block diagram showing a configuration of a motion prediction controller according to the present invention, and FIG. Fig. 4 is a block diagram showing the configuration of the sub-pixel motion prediction unit according to the present invention.

Referring to the drawings, a high-performance H.264 / AVC encoder motion prediction apparatus 1 according to the present invention includes a motion prediction controller 100, a refiner motion predictor 200, a subpixel motion predictor 300, and a motion vector. RAM 400 may be configured to include.

The motion prediction controller 100 may set an initial search point by calculating an intermediate motion vector value mvp from motion vectors of three neighboring macroblocks for the current macroblock.

The motion prediction control unit 100 may include an intermediate motion vector value generation module 110 that calculates an intermediate motion vector value mvp by using the magnitude and direction of motion vectors of three neighboring macroblocks with respect to the current macroblock, Memory control for receiving the current macroblock data and the reference frame data from the external memory and the search point mode selection module 120 for setting the search point mode and the initial search point according to the direction of the intermediate motion vector. Module 130 may be included.

The refinery motion prediction unit 200 may generate a refinery motion vector for the current macroblock by using an initial search point set through the motion prediction control unit 100.

The refinery motion prediction unit 200 compares the calculated SAD value to the SAD calculation module 210 for calculating a SAD (Sum of Absolute Difference) value for each initial search point set by the motion prediction controller 100, and then calculates a minimum SAD value. Is stored for each macroblock size and generates a refiner motion vector for the macroblock having the minimum SAD value selected by the minimum value selection module 220 and the minimum value selection module 220 for selecting the macroblock having the minimum SAD value. It may include a motion vector generation module 230.

Also, the subpixel motion predictor 300 may generate a subpixel motion vector for the current macroblock by using pixels of a reference frame corresponding to the refinement motion vector generated by the refiner motion predictor 200.

The subpixel motion predictor 300 receives an input pixel of a reference frame corresponding to a refiner motion vector and interpolates in quad pixel units to generate 16 × 16 quad pixel data, and generates an interpolation module 310. A SAD calculation module 320 that calculates SAD values by performing a SAD operation on 1 × 16 quad pixel data and stores the calculated SAD values, and stores a subpixel of a macroblock having a minimum SAD value among the calculated SAD values. It may include a subpixel motion vector generation module 330 for generating a motion vector. In addition, the subpixel motion vector generated by the subpixel motion predictor 300 is stored in the motion vector RAM 400.

In addition, the SAD calculation modules 210 and 320 included in the refinery motion predictor 200 and the subpixel motion predictor 300 to perform the SAD operation may be configured in the same form.

5 is a flowchart illustrating a motion prediction method for a high performance H.264 / AVC encoder according to the present invention, FIG. 6 is a flowchart illustrating a method of generating a refiner motion vector for a current macroblock according to the present invention, and FIG. A flowchart illustrating a method of generating a subpixel motion vector for a current macroblock according to the present invention. Next, a motion prediction method for a high performance H.264 / AVC encoder according to the present invention will be described with reference to FIGS.

First, as shown in FIG. 8, the motion prediction controller 100 calculates an intermediate motion vector value mvp using the magnitude and the direction of the motion vectors of three neighboring macroblocks with respect to the current macroblock (S101).

The intermediate motion vector value generation module 110 may calculate the intermediate motion vector value mvp through the following relational expression.

[Formula 1]

Here, MV _left is the motion vector of the left macroblock with respect to the current macroblock, MV _top is the motion vector of the upper macroblock with respect to the current macroblock, and MV _{top - right} is the motion of the upper-right macroblock with respect to the current macroblock. Represents a vector.

Thereafter, the search point mode selection module 120 sets the search point mode by comparing the x coordinate and the y coordinate of the intermediate motion vector value generated by the intermediate motion vector value generation module 110 and sets the search point mode to the direction of the intermediate motion vector value. Accordingly, an initial search point may be set (S102).

[Formula 2]

In particular, as shown in FIG. 9, the search point mode selection module 120 may set the search point mode to nine modes for single layer motion prediction in order to reduce a large amount of computation according to the hierarchical motion prediction.

Such single layer motion prediction does not require the control logic according to the hierarchical motion prediction, reduces the computational amount with a small search point, and sets the initial search point with the direction of the intermediate motion vector value. There is a feature that can have a large search area.

For example, when there is no motion vector of the neighboring macroblock, the search point mode is set as shown in FIG. 9 (a), and when the intermediate motion vector value has the linear direction, FIGS. 9 (b) to (e) As shown in FIG. 9, if the intermediate motion vector value has a diagonal direction, the search point mode may be set to the rectangular search pattern as illustrated in FIGS. 9 (f) to (i).

In addition, the set initial search point may be set to at least five to nine initial search points by moving the intermediate motion vector value according to the search point mode, that is, the direction of the intermediate motion vector value.

Subsequently, a refinement motion vector is generated for the current macroblock based on the initial search point set in the refinery motion prediction unit 200 (S102). In this case, the SAD calculation module 210 may calculate the SAD value for each initial search point through the following relational expression (S201).

[Formula 3]

Here, F _t represents a current frame and F _{t -i} represents a reference frame.

In particular, the SAD calculation module 210 according to the present invention is composed of 16 4 × 4 SAD modules, as shown in FIG. 10, to obtain the difference between the current pixel and the reference pixel in two cycles, and to determine the size of each macroblock. SAD values for one search point may be calculated by adding the difference values corresponding to.

The minimum value selection module 220 compares the calculated SAD values, stores the minimum SAD value for each macroblock size, and selects a macroblock having the minimum SAD value (S202). The refinery motion vector generation module 230 may generate a refinery motion vector IMV of a macroblock having a minimum SAD value (S203).

In this case, one motion vector is generated when the macroblock size having the minimum SAD value is 16 × 16, and 16 motion vectors are generated when the size of the macroblock having the minimum SAD value is 4 × 4.

Next, the subpixel motion prediction unit S300 may generate a subpixel motion vector for the current macroblock using the generated refinement motion prediction vector (S104).

The subpixel motion vector is generated by receiving a pixel of a reference frame corresponding to the refinement motion vector generated by the interpolation module 310, and outputting a half pixel corresponding to the width in 9 cycles, as shown in FIG. 11. Outputs 8 x 8 half pixels for sub-pixel motion prediction for 12 cycles.

The output 8 × 8 half-pixels are input to the 2-tap filter to interpolate in quad pixel units, and output 16 × 16 quad pixel data using four modules that perform 4 × 4 quad pixel interpolation (S301). ).

16 × 16 quad pixel data is input to the SAD calculation module 320 to perform a SAD operation for one cycle to calculate a SAD value (S302). The subpixel motion vector generation module 330 may store the calculated SAD value and generate a subpixel motion vector having a minimum SAD value (S303).

Peak Signal to Noise Ratio (PSNR) results simulating motion prediction in accordance with the present invention and comparative techniques, for example Enhanced Predictive Zonal Search, Tourapis H.-YC, Tourapis AM, "Fast motion estimation within the H. 264 codec, "Proceedings of Multimedia and Expo ICME, vol. 3, PP.517-520).

As shown in Table 1, it can be seen that the PSNR of the present invention is reduced by 0.061 dB on average compared to EPZS.

[Table 1]

In addition, as shown in Table 2, in the present invention, the maximum operating frequency is 166 MHz and the number of gates is 217.92 k, which is reduced by 66.08 k compared with the prior art.

[Table 2]

In addition, according to the present invention, it takes 312 cycles to encode one macroblock, which can be confirmed to be approximately 69% improved compared to the conventional method.

As described above, according to the present invention, by predicting the initial search point according to the intermediate value and the direction of the motion vectors of the neighboring macroblocks in motion estimation, the amount of computation is reduced to a small number of search points, and the motion vectors of the neighboring macroblocks are used. Therefore, even in HD resolution, PSNR loss is small compared to a small amount of calculation.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed embodiments, but, on the contrary, is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims. will be. Accordingly, the true scope of the present invention should be determined by the technical idea of the appended claims.

1: motion prediction apparatus 100: motion prediction controller
110: intermediate motion vector value generation module 120: search point mode selection module
130: memory control module 200: purification plant motion prediction unit
210: SAD calculation module 220: minimum value selection module
230: refiner motion vector generation module 300: subpixel motion prediction ㅂ
310: interpolation module 320: SAD calculation module
330: subpixel motion vector generation module

Claims (21)

In the motion prediction device for high performance H.264 / AVC encoder,
A motion prediction controller configured to set an initial search point by calculating an intermediate motion vector value mvp from motion vectors of three neighboring macroblocks for the current macroblock;
A refiner motion predictor configured to generate a refiner motion vector for the current macroblock using the initial search point; And
And a subpixel motion predictor configured to generate a subpixel motion vector for a current macroblock using pixels of a reference frame corresponding to the refiner motion vector generated by the refiner motion predictor.
The motion prediction control unit,
An intermediate motion vector value generation module for calculating an intermediate motion vector value mvp using the magnitude and the direction of the motion vectors of three neighboring macroblocks with respect to the current macroblock;
A search point mode selection module configured to set a search point mode and an initial search point according to the direction of the intermediate motion vector value; And
And a memory control module configured to receive data of a current macroblock and data of a reference frame from an external memory and store the data in an internal memory.
The search point mode selection module sets a search point mode using a diamond search pattern when the intermediate motion vector value has a linear direction and a rectangular search pattern when the intermediate motion vector value has a diagonal direction.
delete The method of claim 1,
The intermediate motion vector value generation module calculates an intermediate motion vector value using the following relational expression.

Where MV _left is the motion vector of the left macroblock for the current macroblock, MV _top is the motion vector of the upper macroblock for the current macroblock, and MV _top-right is the _top-right macroblock of the current macroblock. Represents a motion vector)
The method of claim 1,
The search point mode selection module sets the search point mode by comparing the x coordinate and the y coordinate of the intermediate motion vector value through the following relational expression.

delete The method of claim 1,
The search point mode selection module sets the search point mode in nine modes according to the direction of the intermediate motion vector value.
The method according to claim 6,
The search point mode selection module sets an initial search point to at least five to nine search points according to the set search point mode.
The method of claim 1,
The refinery motion prediction unit,
A SAD calculation module for calculating a sum of absolute difference (SAD) value for each initial search point;
A minimum value selecting module for comparing the calculated SAD values, storing a minimum SAD value for each macroblock size, and selecting a macroblock having a minimum SAD value; And
And a refiner motion vector generation module for generating a refiner motion vector for the macroblock having the minimum SAD value selected by the minimum value selecting module.
The method of claim 8,
The SAD calculation module calculates a SAD value through the following relational expression.

Where F _t represents the current frame and F _{t -i} represents the reference frame
The method of claim 8,
The refinery motion vector generation module,
When the size of the macroblock having the minimum SAD value is 16 × 16, one motion vector is generated.
And 16 motion vectors are generated when the size of the macroblock having the minimum SAD value is 4x4.
The method of claim 1,
The subpixel motion prediction unit,
An interpolation module for generating 16 × 16 quad pixel data by receiving pixels of a reference frame corresponding to the refiner motion vector and interpolating in quad pixel units;
A SAD calculation module configured to calculate a SAD value by performing a SAD operation on the generated 16 × 16 quad pixel data for one cycle;
And a subpixel motion vector generation module for storing a calculated SAD value and generating a subpixel motion vector of a macroblock having a minimum SAD value among the calculated SAD values.
In the motion prediction method for high performance H.264 / AVC encoder,
A first step of the motion prediction controller calculating an intermediate motion vector value mvp by using the magnitude and the direction of the motion vectors of three neighboring macroblocks with respect to the current macroblock;
A second step of setting, by the motion prediction controller, a search point mode and an initial search point from the calculated intermediate motion vector value;
A third step of generating a refinery motion vector for the current macroblock based on the initial search point set in the second step; And
And a fourth step of generating, by the subpixel motion predictor, the subpixel motion vector for the current macroblock using the refinement motion prediction vector generated in the third step.
In the second step, the initial search point is set to at least five to nine search points according to the search point mode is set.
13. The method of claim 12,
In the first step, the intermediate motion vector value (mvp) is calculated through the following relational expression.

Where MV _left is the motion vector of the left macroblock for the current macroblock, MV _top is the motion vector of the upper macroblock for the current macroblock, and MV _{top - right} is the _{top - right} macroblock of the current macroblock. Represents a motion vector)
13. The method of claim 12,
In the second step, the search point mode is set by comparing the x coordinate and the y coordinate of the intermediate motion vector value through the following relational expression.

15. The method of claim 14,
The search point mode,
If the intermediate motion vector has a linear direction, a diamond search pattern is used.
And a rectangular search pattern is used if the intermediate motion vector has diagonal direction.
The method of claim 15,
The search point mode is set to nine modes, the motion prediction method.
delete 13. The method of claim 12,
In the third step,
Calculating a SAD value for each initial search point by a sum of absolute difference (SAD) calculation module;
Comparing a calculated SAD value by a minimum cost selection module to store a minimum SAD for each macroblock size, and selecting a macroblock having a minimum SAD value; And
And generating, by the refiner motion vector generation module, a refiner motion vector (IMV) of the macroblock having the minimum SAD value.
19. The method of claim 18,
The step of calculating the SAD value is a motion prediction method, characterized in that performed through the following equation.

Where F _t represents the current frame and F _{t -i} represents the reference frame
19. The method of claim 18,
The process of generating the refinery motion vector,
When the size of the macroblock having the minimum SAD value is 16 × 16, one motion vector is generated.
When the size of the macroblock having the minimum SAD value is 4 × 4, 16 motion vectors are generated.
13. The method of claim 12,
The fourth step,
An interpolation module receiving 16 pixels of the reference frame corresponding to the refinement motion vector generated in the third step and interpolating in quad pixel units to generate 16 × 16 quad pixel data;
Calculating a SAD value by performing an SAD operation on the 16 × 16 quad pixel data generated by the SAD calculation module for one cycle; And
And a subpixel motion vector generation module storing the calculated SAD value and generating a subpixel motion vector of a macroblock having a minimum SAD value.

Images