KR20070076672A - Algorithm for motion estimation and block partition mode - Google Patents

Abstract

An algorithm for motion estimation and block partition mode decision is provided to suggest a sub-pel search order capable of calculating an SAD(Sum of the Absolute Difference) and a cost every cycle without delay. A motion vector of an integer-pel unit is searched by using information about a current macro-block and macro-blocks within a search area of a previous frame, and a block partition mode is decided while the motion vector is searched(210). It is determined whether a motion vector of a sub-pel unit is additionally searched at the decided block partition mode, and a motion vector of the sub-pel unit is searched in accordance with a result of the determination. When the motion vector of the integer-pel unit is searched and the block partition mode is decided, a plurality of sub-blocks is partitioned into a plurality of groups and a motion vector is decided by the partitioned groups, wherein a block partition mode of the groups is decided if a motion vector for at least two of the groups is decided.

Description

Algorithm for motion estimation and block partition mode

BRIEF DESCRIPTION OF THE DRAWINGS In order to better understand the drawings cited in the detailed description of the invention, a brief description of each drawing is provided.

1 shows a part of a compressor that performs motion estimation and mode determination for a video.

2 is a signal flowchart of a fast algorithm for motion estimation and block division mode determination according to an embodiment of the present invention.

3 shows an example of a finite state machine (FSM).

FIG. 4 shows obtaining a SAD value between a 4x4 block of the current macroblock and a 4x4 block of the previous macroblock.

5 shows a procedure for calculating the cost of a 4x4 block in a macroblock.

FIG. 6 illustrates a step and method for calculating the cost of each of the sixteen macroblocks shown in FIG. 5.

7 illustrates a data path of a reuse register file according to another embodiment of the present invention.

8 is an example of a register reuse lookup table.

9 shows the arrangement relationship of the integer pels and the half pels.

10 illustrates a purified water pel search area and a cell pel search area according to another embodiment of the present invention.

11 is a layout view of the clean water pel, half pel and quarter pel.

It is a pel arrangement figure for demonstrating the interpolation method of obtaining 1/2 pel.

The present invention relates to video compression, and more particularly, to a motion estimation and block partition mode determination algorithm applicable to H.264.

Delivering digital video from a source (camera or stored video) to a destination (display) requires a key process called compression (encoding) and reconstruction (decoding). Through the above process, 'raw' digital video, which is a burden of bandwidth, is compressed into a manageable size for transmission or storage, and then restored for display.

This compression and decompression process was established by the international standard ISO / IEC 13818, known as MPEG-2, for the digital video industry such as digital television broadcasting and DVD-Video. This was developed: MPEG-4 Visual and H.264. The H.264 standard is a video compression technology standard established by the International Telecommunication Union (ITU-T) and is also called MPEG-4 Moving Picture Experts Group-Phase 4 Advanced Video Coding (AVC). The two compression standards have gone through the same process and have some common features, but with very different goals.

The goal of MPEG-4 Visual is to overcome the limitations of relying only on rectangular video images, providing a free and flexible structure for video communications, enabling the use of the best features such as efficient video compression and object-based processing. In contrast, H.264 has a more practical goal: to perform rectangular video compression as in previous standards in a more efficient, powerful and practical way, such as broadcast, storage, and streaming, which are widely used in the market. To support the field.

As in the conventional video encoding method, the H.264 standard generates a predictive signal that estimates motion from an already encoded video frame, and discrete cosine transforms the difference signal (residual signal) between the video frame and the predictive signal. Cosine Transform (hereinafter referred to as DCT) is based on a Motion Estimation DCT technique. The H.264 standard has two to three times better compression rate than the existing MPEG2. For example, even if the standard definition (SD) service of 4 Mbps (Mega bit per second) level in MPEG2 is applied, the H.264 standard is 1.5Mbps. Is enough. Since the H.264 standard can deliver DVD-quality high-quality video at speeds below 1 Mbps, satellite, cable and Internet Protocol TV (IPTV) operators are also interested in H.264.

1 shows a part of a compressor that performs motion estimation and mode determination for a video.

 Referring to FIG. 1, the compressor 100 includes a memory 110, a first control block 120, a second control block 130, a BSU 140, an MV control block 150, a mode and The MV decision block 160 and the main control block 170 are provided.

The memory 110 includes a first memory 111 that stores pixel data of a current macro-block and a second memory 112 that stores pixel data of a reference search area. do. The first control block 120 reads the current macro block in order of 4 × 4 from the first memory 111 in response to the control signal of the main control block 170. The second control block 130 reads the macro blocks of the reference search area in the order of 4x4 from the first memory 111 in order in response to the control signal of the main control block 170, and sub-pels. In the sub-pel mode, the sub-pel interpolation performs the role of a filter. Here, pel is short for pixel.

The Basic Search Unit (BSU) 140 includes a SAD (Sum of 4 × 4 unit) of the current macro block and the reference search area received from the first control block 120 and the second control block 130 with respect to the macro block. Calculate the Absolute Difference. The MV control block 150 receives a motion vector of blocks around the current macro block and generates a Predicted Motion Vector (P-MV) that estimates the motion vector of the current macro block. The mode decision block 160 receives the P-MV from the SAD and the MV prediction block 150 calculated from the BSU 140 to determine an optimal MV and block partition mode. The main control block 170 controls the operations of the first control block 120 and the second control block 130 according to the internal information of the mode determination block 160.

Currently, a lot of research is being conducted on the H.264 motion estimation (ME), and the H.264 motion estimation has the following problems.

1. Due to the complexity of the H.264 standard, the hardware that implements it is large.

2. Excessive registers are needed to store intermediate calculations for all modes and motion vectors.

3. Increased memory access to read data from other nearby pels for subpel search.

In order to overcome the above problems,

1. Development of Fast Algorithm for Motion Estimation and Block Partition Mode Decision,

2. a reduction in the number of registers used to store interpolated result values for block division mode and motion vectors,

3. We propose differentiation between sub pel search area and integer pel search area.

An object of the present invention is to provide a fast algorithm for motion estimation and block division mode determination.

Another technical problem to be achieved by the present invention is to provide a method for differentiating the purified water pel search range and the sub-pel search range.

Another technical problem to be achieved by the present invention is to provide a sub-pel search sequence in which SAD or cost can be obtained every cycle without delay.

According to an aspect of the present invention, a motion estimation and block division mode determination algorithm includes an integer pel unit motion vector search, a block partition mode determination step, and a sub pel unit motion vector search step.

In the step of determining the integer pel motion vector and determining the partition mode, the motion vector of the integer unit is searched using information on the macroblocks in the search area of the current macroblock and the previous frame, and the block is in the middle of the search. Determine the split mode. In the subpel unit motion vector search step, in the block partition mode determined in the integer unit unit vector search and block division mode determination step, it is determined whether the motion vector of the subpel unit should be further searched, and according to the determination result. Search for motion vectors in subpel units.

DETAILED DESCRIPTION In order to fully understand the present invention, the operational advantages of the present invention, and the objects achieved by the practice of the present invention, reference should be made to the accompanying drawings which illustrate preferred embodiments of the present invention and the contents described in the drawings.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. Like reference numerals in the drawings denote like elements.

2 is a signal flowchart of a fast algorithm for motion estimation and block division mode determination according to an embodiment of the present invention.

Referring to FIG. 2, the fast algorithm includes an integer pel unit motion vector search and block division mode determination step 210, a sub pel unit motion vector search and block division mode determination step 230, and a motion vector and block division mode. A final decision step 250 is provided.

The integer pel unit motion vector search and block division mode determination step 210 may search for a motion vector of an integer unit by using information about macroblocks in a search region of a current macroblock and a previous frame. To determine the block division mode.

The subpel unit motion vector search and block division mode determination step 230 includes a skip determination step 235 and a subpel search step 240.

The skip determination step 235 includes a first determination step 236, a first skip determination step 237, a second skip determination step 238, and an area deviation determination step 239.

The first determination step 236 determines whether the block division mode determined in the integer pel unit motion vector search and block division mode determination step 210 is 16 × 16.

In the first skip determination step 237, when the block division mode determined in the integer pel unit motion vector search and block division mode determination step 210 is 16 × 16, the SAD for the predicted motion vector (P-MV) is determined. The value of Sum of the Absolute Difference (SAD16 × 16 (PMV)) is compared with the Skip Threshold Value (Th16 _LL ), where the predicted motion vector (P-MV) is already calculated for the peripheral motion. It can be easily obtained from a vector without any special calculation.

In the second skip determination step 238, the block division mode determined in the integer pel unit motion vector search and block division mode determination step 210 is not 16 × 16 or is determined for the predicted motion vector P-MV. Mean of the Absolute Difference (MAD) and SPME skip reference values calculated for the block division mode when the value of SAD (SAD16 × 16 (PMV)) is not less than or equal to the skip reference value (Th16 _LL ). (Sub Pel Motion Estimation Skip Threshold Value, Thb).

In the region deviation determining step 239, when the MAD value is not equal to or smaller than the SPME skip reference value Thb, it is determined whether the SPME search region is included in the range of the distinguished region. See the descriptions of FIGS. 10 and 10 for the distinction area.

In the sub-pel search step 240, when the SPME search area is not included in the distinguished area as determined in the area deviation determining step 239, the sub-pel search step 240 searches.

Here, in the three cases listed below, the subpel search 240 is not performed.

1. When the skip reference value Th16 _LL is greater than or equal to the value of the predicted motion vector P-MV, as determined in the first skip determination step 237.

2. The SPME skip reference value Thb is greater than or equal to the MAD value as a result of the determination in the second skip determination step 238.

3. The SPME search region is included in the distinguished region as a result of the determination in the region deviation determining step 239.

The final determination step 250 of the motion vector and the block division mode may include information on the subpels obtained by performing the subpel search step 240 in addition to the three cases determined as not performing the subpel search 240. When received, the final motion vector is determined, and at the same time, MVd is calculated by calculating a predicted motion vector (P-MV) according to a predetermined mode for each partition block, and finally, a partition partition mode (Final Partition mode) Is determined. Here, MVd means a difference value between MV and P-MV (= MV-P-MV) and is transmitted to a functional block (not shown) that performs variable length coding (VLC). The skip reference value Th16 _LL and the SPME skip reference value Thb used above may be arbitrarily determined by computer simulation or experience.

Hereinafter, a signal flowchart for an algorithm for motion estimation and block division mode determination according to an embodiment of the present invention shown in FIG. 2 will be described.

The block division mode mentioned in the integer pel unit motion vector search and block division mode determination step 210 will be described.

The motion vector for the current macroblock is basically determined for a 4x4 block, which in some cases is one of 8x4, 4x8, 8x16, 16x8, and 16x16 macroblocks. Can be determined. As described above, it is the above-described block division mode decision to determine which type of macro block the motion vector corresponds to. The block division mode determination may further include a skip mode for not selecting any type of the plurality of macroblocks and determining a motion vector.

When determining the block division mode and the motion vector, the cost concept expressed in Equation 1 is used.

Cost = SAD + λ × Rmv + α

Where λ is a Lagrangian multiplier, a function of Qp of H.264, and Qp is a quantization coefficient. Rmv is a motion vector coding bit rate, and a motion vector closer to the predicted motion vector (P-MV) has a smaller value. α is a weight that can be assigned autonomously by the user and is generally an integer.

In the integer pel unit motion vector search and block division mode determination step 210, a motion vector corresponding to the lowest cost value shown in Equation 1 is determined and the block division mode corresponding thereto is preliminarily determined. do. The motion vector determined at this time is determined by motion estimation in integer units.

3 shows an example of a finite state machine (FSM).

The motion estimation (ME) and the mode determination (MD) applying the FSM shown in FIG. 3 are controlled by the main control block 170. The FSM operates in macroblock units and is performed in a plurality of hierarchical layers.

In each layer, SAD values are calculated in units of 4 × 4, and the SAD values are made in the BSU 140.

3 is an example of the hierarchical motion search method. The present invention can be applied to other integer pel unit motion search methods in addition to the above method.

FIG. 4 shows obtaining a SAD value between a 4x4 block of the current macroblock and a 4x4 block of the previous macroblock.

4, the absolute value (| C-P |) of the difference between the current (C) macroblock and the previous (P) macroblock is shown. The number indicated by the subscript of the alphabet displayed inside the cell of each macroblock corresponds to the cell position of the macroblock. The absolute values of the differences of the cells are obtained, and these are added step by step to finally calculate the value of SAD (4 × 4).

When searching for a 4 × 4 unit motion vector, in order to obtain the SAD having the smallest value, the distance from the predicted motion vector (P-MV) of the corresponding macroblock is compared with the same SAD value in the search group. A SAD and a motion vector MV that are close to the predicted motion vector P-MV are selected.

5 shows a procedure for calculating the cost of a macroblock.

Referring to FIG. 5, when four blocks are grouped into one group, 16 subblocks are divided into four groups, and the cost of each macroblock constituting any one group is calculated in the direction of the letter Z. The four blocks grouped into one group are 0-3, 4-7, 8-11 and 12-15.

FIG. 6 illustrates a step and method for calculating the cost of each of the sixteen subblocks shown in FIG. 5.

Referring to FIG. 6, the cost of each of the subblocks 4x4_0, 4x4_1, 4x4_2, and 4x4_3 included in the first group is calculated in the Z direction. First, the costs Cost4_0 and Cost4-1 are calculated in order for the two subblocks 4x4_0 and 4x4_1. The two costs (Cost4_0 and Cost4-1) are used to calculate the cost (Cost84_0) for the 8x4 mode, which can be formed by combining two subblocks (4x4_0 and 4x4_1). Two subblocks (4x4_0, 4) using the cost (Cost4_2) of the third calculated third subblock (4x4_2) and the cost (Cost4_0) of the first subblock (4x4_0) already calculated The cost Cost48_0 for the 4x8 mode which can be formed by adding up x4_2) is calculated. Two subblocks (4 × 4_2, 4 ×) using the cost (Cost4_3) of the fourth calculated fourth block (4 × 4_3) and the cost (Cost4_2) of the third subblock (4 × 4_2) already obtained. The cost (Cost84_1) for the 8x4 mode that can be formed by combining 4_3) is obtained, and two subblocks (4x4_1, 4) are obtained by using the cost (Cost4_1) of the second subblock (4x4_1). The cost (Cost48_1) for the 4x8 mode that can be formed by combining the x4_3) is obtained, and the cost (Cost88_0) for the 8x8 mode is obtained using all four costs (Cost4_0, Cost4_1, Cost4_2, and Cost4_3). .

The block division mode for the four subblocks included in the first group may be determined using the cost for the four subblocks of the first group calculated according to the above order.

If the same operations performed in the first group are performed on the subblocks 4x4_4, 4x4_5, 4x4_6, and 4x4_7 in the second group, the subblocks included in the second group The block splitting mode for the same may be determined through the same process.

Subblocks included in the third group (4 × 4_8, 4 × 4_9, 4 × 4_10, 4 × 4_11) and subblocks included in the fourth group (4 × 4_12, 4 × 4_13, 4 × 4_14, 4 × 4_15 ), The same operation as the operation performed in the first group and the second group described above may be performed to determine each block division mode.

In particular, the final cost (Cost88_0, Cost88_1) of the first group and the second group can be used to calculate the cost (Cost168_0) for the 16x8 mode, and the final cost (Cost88_2, Cost88_3) of the third and fourth group. The cost Cost168_1 for the 16 × 8 mode can be calculated using. The final cost (Cost88_0, Cost88_2) of the first group and the third group can be used to calculate the cost (Cost816_0) for 8x16 mode, and the final cost (Cost88_1, Cost88_3) of the second and fourth group. The cost Cost816_1 for the 8x16 mode can be calculated.

Conventionally, since the block division mode having the minimum cost is determined after not only the operation for the 16 macroblocks but also the subpel level to be described later, all intermediate calculation result values are selected until the block division mode is selected. Had to save. Therefore, a considerable number of registers are required, which eventually leads to an increase in the size of the hardware.

However, referring to FIG. 6, for the four macroblocks forming the first group of the total 16 macroblocks, the block partition mode may be preliminarily determined without waiting for the operation result of another group. In addition, even for the four macroblocks forming the second group, the block division mode can be independently determined regardless of the block division mode determined in the first group. Subsequently, the third group and the fourth group may also determine the block division mode, respectively, regardless of the block division mode determined in the first and second groups.

With reference to the foregoing and FIG. 6, except for the final costs (Cost88_0, Cost8-1, Cost88_2, Cost8-3) of the group, the cost for each of the four macroblocks included in the group and the calculation are performed using them. It can be seen that the median values of the ones are no longer used in subsequent computations. Accordingly, the present invention proposes to reuse a register that stores intermediate values, except for a register that stores the final cost (Cost88_0, Cost8-1, Cost88_2, Cost8-3) of the group.

7 illustrates a register data path according to another embodiment of the present invention.

Referring to FIG. 7, the register data path includes a register file 710 and a multiplexer 730.

The register file 710 is composed of a total of 12 files including 12 columns of 8-bit registers. The data of the previous frame of the pels (Previous Pels) and the data of the interpolated pels (Interpolated Pels) are input. Save / output SAD values and motion vector values MVs.

The multiplexer 730 determines the information about the motion estimation state (ME state), a processor counter (Process Counter), a macroblock shape signal (mb_type) for determining a macroblock partition size, and a sub macroblock partition size. In response to information such as the sub macroblock type (sub_mb_type) and the like, data of a Pel of the previous frame (Previous Pels), interpolated Pels of data (Interpolated Pels), SAD values (SAD values) and motion vector values (MVs) ) And one of the Pels of the previous frame input from the register file 710, Interpolated Pels of interpolated Pels, SAD values, and motion vector values (MVs). Select to pass to the register file 710.

The values that are input and stored in the registers to be reused as described above are values that are temporarily stored according to each state without being used at the same time. In order to prevent these values from conflicting with each other, the register reuse illustrated in FIG. 8 is reused. You can do this by creating a table.

Hereinafter, a step 230 of estimating the motion of the subpel will be described.

One of the technical ideas proposed by the present invention is that the block division mode determination is performed at the integer pel level, and the search at the sub pel level is performed in the block division mode already determined. Therefore, there is a case where the search at the subpel level does not need to occur, and even when performing the search at the subpel level, the burden of storing all intermediate values such as SAD and cost can be reduced. Can be run at high speed.

That is, in the previously determined block division mode, when performing the subpel search in units of 4 × 4, using the motion vector (N-MV) of the neighboring block which can be known in advance through the previously searched information, the motion vector coding bit The rate Rmv can be calculated efficiently. In addition, since the subpel search is performed in units of blocks divided according to the previously determined block partitioning mode, SAD and cost values to be stored can be minimized.

In H.264, searching for up to 1/4 pel (or quarter pel) units in the luminance component increases the probability of finding the optimal motion vector (MV). The quarter pels in H.264 are output via bi-linear interpolation of half pels (or half pels). In this case, since a 6-tap filter is used to obtain 1/2 pels, three more pels are required to be back and forth or up and down about the 1/2 pel.

9 shows the arrangement relationship of the integer pels and the half pels.

Referring to Figure 9, in order to calculate the 1/2 pel (b) indicated by the arrow in the ellipse, information about three integer pels (black squares), each of which is about 3 1/2 pels (b), is needed. As a result, the search range is increased to ± 3.

In general, information about the pels included in the search area is read from the SDRAM in multiples of 16. However, since the information on the pels outside the area indicated by the dotted line must be read in units of pels in which words are not aligned, the use efficiency of bus bandwidth is greatly reduced. There is this.

Accordingly, in the present invention, the problem is solved by differentiating the search region of the integer pel unit and the search region of the sub pel unit, and it is determined in the departure region determination step 239.

10 illustrates a distinguishing area according to another embodiment of the present invention.

Referring to Fig. 10, the hatched distinction area is an area used for the constant pel search but not for the sub pel search. In other words, the subpel search is skipped in the distinguished area. In the present invention, it is ± 16 for the integer pel search, but ± 13 for the subpel search, so that all ± 3 pels for 6-tap filtering the 1/2 pel are placed within the integer pel search area of ± 16. Increasing memory bandwidth.

For pels of ± 14, ± 15, and ± 16, large motion is expected, so the corresponding motion vector will generally be large. In this region, it is generally known that there is almost no difference in the image quality compressed and restored by retrieving and reproducing the water purge only through the search and further including the subpel search. In addition, based on the fact that the motion vector (MV) near the boundary of the search area is generally not large in one screen, it is proposed not to perform the operation on the subpel in the distinct area. In the present invention, it is determined in the region deviation determination step 239. Even if the subpel search area is reduced by the distinguished area, the image quality has little effect on the image quality, and thus a detailed description thereof will be omitted since a person skilled in the art can easily verify through computer simulation.

By reducing the subpel search area as described above, the size of the memory 110 to store the intermediate result value of the operation during the subpel search can be reduced and the amount of calculation is also reduced.

In addition, in order to reduce the execution time of the motion search in the sub-pel, the MAD is calculated for each block divided for the mode determined in the integer pel, and the values are not compared with the skip reference value (THb) so that the SPME may not be performed (237). .

The present invention also proposes a motion search sequence in the subpel, which is performed in the subpel search step 240 so that the SAD and the cost can be obtained every clock.

11 is a layout view of the clean water pel, half pel and quarter pel.

Referring to FIG. 11, the portions marked in black (IP, NW, N, NE, WN, EN, W, E, WS, ES, S, SE) are integer pels, and the portion starting with H is 1/2. Pel, quarter starting with F.

To get the cost without time delay every clock cycle,

1. Search in the order of 1/2 pellets (H3, H4), 1/4 pellets (F18, F19, F20, F21) and the purified water pellet (IP) in the same row as the purified water pellet (IP) ,

2. Among the remaining 1/2 pels disposed around the purified water pel IP, the 1/2 pels (H0, H1, H2) at the top of the purified water pel (IP) and the half pels at the bottom ( H5, H6, H7) in order,

3. The order of the fourth quarters at the bottom (F0, F1, F2, F3, F4, F5, F6, F7, F8, F9, F10, from the remaining quarters at the top of the integer fel) (IP) F11, F12, F13, F14, F15, F16, F17, F22, F23, F24, F25, F26, F27, F28, F29, F30, F31, F32, F33, F34, F35, F36, F37, F38, f39) Navigate to.

If the subpels are searched in the above order, the cost of the subpels can be extracted at every cycle of the clock, thereby making the use of the clock more efficient. The smallest search point is selected as the final motion vector by investigating the cost of the subpel within the area of ± 0.75 centered on the integer pel (IP).

It is a pel arrangement figure for demonstrating the interpolation method of obtaining 1/2 pel.

Referring to FIG. 12, a six-tap Finite Impulse Response (FIR) filter is used to obtain 1/2 pels located between hatched water purification pels. For example, 1/2 pel b is calculated from equation (6) from six horizontal integer pels (E, F, G, H, I, J).

b = round ((E-5F + 20G + 20H -5I + J) / 32)

Likewise, 1/2 pel h is interpolated by filtering A, C, G, M, R, T.

Once all 1/2 pels have been found, find the 1/4 pel located between the purified pelvis and the 1/2 pel. The 1/4 pel is obtained by interpolating the purified water pels and the 1/2 pels on their left, right, or top and bottom sides. However, as shown in Equation 2, since the bit corresponding to the value of the sub pel b is quite large, when consuming a quarter pel using this it consumes a lot of power.

In the present invention, in the sub-pel search step 240, to obtain the quarter pel, the result value of the operation expressed in Equation 2 is not used as it is, but clamped or normalized to be used. Suggest that. The fact that the value of the quarter pel obtained by clamping or normalizing the value of the half pel and the quarter pel obtained by using the value of the half fel that is not clamped or normalized is almost the same. This can be easily confirmed through computer simulation. Therefore, when clamping or normalizing the value of the 1/2 pel, it is possible to obtain the effect of maximally reducing the amount of calculation without affecting the image quality to be reproduced.

As described above, optimal embodiments have been disclosed in the drawings and the specification. Although specific terms have been used herein, they are used only for the purpose of describing the present invention and are not intended to limit the scope of the present invention as defined in the claims or the claims. Therefore, those skilled in the art will understand that various modifications and equivalent other embodiments are possible therefrom. Therefore, the true technical protection scope of the present invention will be defined by the technical spirit of the appended claims.

As described above, the motion estimation and block division mode determination algorithm according to the present invention has an advantage of enabling hardware to be operated at high speed and low power without much deterioration in image quality during reproduction.

Claims (15)

Integer pel unit motion vector search and block partition mode that searches for a motion vector in integer units using information on macroblocks in a search region of a current macroblock and a previous frame, and determines a block partition mode during the search. Determination step; And In the block division mode determined in the step of searching the motion vector of the integer pel unit and determining the block division mode, it is determined whether the motion vector of the sub pel unit should be further searched, and the motion vector of the sub pel unit is searched according to the determination result. Motion estimation and block division mode determination algorithm. The method of claim 1, wherein the integer pel unit motion vector search and block division mode determination step, When the size of the current subblock is 4 × 4, the block partitioning mode determined is one block division among 8 × 4, 4 × 8, 8 × 16, 16 × 8, and 16 × 16 block division types. A motion estimation and block division mode determination algorithm according to a shape. The method of claim 2, wherein the integer pel unit motion vector search and the block division mode determination step include: The plurality of subblocks are divided into a plurality of groups and a motion vector is determined for each of the divided groups. In this case, if a motion vector for at least two divided groups is determined, the block division mode between the groups is determined. Motion estimation and block division mode determination algorithm. The method of claim 3, wherein the integer pel unit motion vector search and block division mode determination step, And a portion of the storage device used in the process of determining the motion vector of one of the divided groups is reused in the process of determining the motion vector of the other group. The method of claim 4, wherein the integer pel unit motion vector search and block division mode determination step include: Reuse Lookup Table with Reusable Storage Locations, General Information on Pellets, and Sub Pel Motion Estimation (SPME) to ensure that data from reusable storage does not conflict with each other. A motion estimation and block division mode determination algorithm, characterized in that for referring to. The method of claim 1, wherein the subpel unit motion vector search step comprises: A skip determination step of determining whether to perform subpel unit motion vector search; And And a subpel search step of searching for a subpel motion vector, when the skip determination step determines that a subpel motion vector is to be searched. The method of claim 6, wherein the skip determination step, A first determination step of determining whether the block division mode determined in the integer pel unit motion vector search and block division mode determination step is 16 × 16; When the block division mode determined in the integer pel unit motion vector search and block division mode determination step is 16 × 16, a skip reference value of SAD (Sum of the Absolute Difference) for the predicted motion vector (P-MV) is skipped. A first skip determination step of comparing a skip threshold value (Th16 _LL ); The block division mode determined in the integer pel unit motion vector search and block division mode determination step is not 16 × 16 or the value of SAD for the predicted motion vector (P-MV) is skipped as a result of the determination in the first skip determination step. Compares the Mean of the Absolute Difference (MAD) value and the SPME Skip Threshold Value (Thb) calculated for the block division mode when it is not smaller than or equal to the reference value (Th16 _LL ). A second skip determination step; And When the MAD value is not less than or equal to the SPME skip reference value (Thb), and the area deviation determination step of determining whether the SPME search area is included in the range of the distinction area, The distinction area is a difference area between the integer pel unit search area and the sub pel unit search area, And a subpel search step is performed if the SPME search area is not included in the distinguished area. The method of claim 7, wherein The integer pel unit search region is larger than the sub pel unit search region, the motion estimation and block partition mode determination algorithm. The method of claim 8, wherein the distinction region, A motion estimation and block division mode determination algorithm having a width of three integer pel units. The method of claim 6, wherein the subpel search step, The data for 1/2 pel is obtained by applying the data for three integer pels in the top, bottom, left and right sides of the corresponding half pel to the equation recommended in H.264 standard. And a half pel data obtained by the above equation is clamped or normalized to reduce its size, and then used to generate data for a quarter pel. The method of claim 6, wherein the subpel search step, First, search in the order of 1/2 pels, 1/4 pels, said purified water pel in the same row as the purified water pel, Subsequently, the remaining 1/2 pels located around the water purification pel are searched for, Finally, the motion estimation and block division mode determination algorithm characterized by searching the remaining quarter pels. The method of claim 11, When searching for the remaining 1/2 pels located in the vicinity of the purified water pel, search in the order of 1/2 pels in the upper part and 1/2 pels in the bottom, And searching for the remaining quarter pels in order from the quarter pels at the top of the integer pels to the quarter pels at the bottom. The method of claim 12, The search for the subpels is performed on the subpels within a range of ± 0.75 around the integer pels when the distance between the integer pels is 1, and the motion estimation and block division mode determination algorithm. The method of claim 1, A final motion vector is determined in response to a result of the subpel unit motion vector search step, and at the same time, a predicted motion vector according to a predetermined block partition mode is calculated for each partition block, and the final motion vector and the predicted motion vector are calculated. And a motion vector and a block division mode final decision step of calculating a difference value and determining a final block division mode. The method of claim 14, wherein the final motion vector, the last block division mode, and the difference value are: A motion estimation and block division mode determination algorithm, which is delivered to a functional block that performs variable length coding (VLC).

Images