KR101036137B1 - Multiple Reference Frame Selection Method for Video Coding - Google Patents

Abstract

The present invention relates to a method of selecting multiple reference frames for video encoding. The multi-reference frame selection method for video encoding according to the present invention comprises the steps of: (a) performing motion estimation on a P16 × 16 block size of a current macroblock to be encoded for a previous frame; (b) calculating temporal complexity from optimal reference frame information of the selected reference region through the motion estimation; And (c) adaptively selecting a candidate reference frame of the current macroblock according to the temporal complexity.

Accordingly, the complexity of motion estimation can be greatly reduced by appropriately selecting the number of reference frames used for mode estimation based on the correlation of the temporal complexity of the current frame and the previous frame.

Motion Estimation, Multiple Reference Frames, H.264

Description

Multiple Reference Frame Selection Method for Video Coding

The present invention relates to a method of selecting multiple reference frames for video encoding, and more particularly, to a method of appropriately selecting candidate reference frames in video encoding using multiple reference frames.

Recently, the compression transmission and encoding technology of digital video is rapidly progressing with the development of communication technology such as broadcasting and internet. In particular, various specifications have been proposed in the field of compression or encoding of digital video, and various efforts have been made to improve performance.

New video coding jointly established by the Joint Video Team (JVT), recently formed by the ITU-T Video Coding Experts Group (VCEG) and the ISO / IEC Motion Picture Expert Group (MPEG). As a standard, H.264 / AVC has been proposed.

H.264 applies a variety of new coding techniques to achieve high coding performance compared to previous standards such as MPEG-2, MPEG-3, and H.263. For example, methods such as quarter-pel accuracy motion prediction, use of multiple reference frames, coding using a variable block size, optimization of rate distortion, and the like have great advantages in inter-screen coding. These coding schemes show an excellent improvement in terms of rate-distortion performance compared to previous coding standards, but also have a disadvantage in that complexity increases excessively.

In particular, the H.264 / AVC standard uses several frames as reference images, as shown in FIG. This shows excellent performance when the movement occurs repeatedly or when an area is hidden by other objects, and the compression efficiency can be maximized.

In addition, unlike conventional video coding standards for motion prediction only in 16 × 16 or 8 × 8 size, as shown in FIG. 2, the H.264 / AVC standard makes the size of the block more varied to 16 × 16. The motion prediction is performed in units of 16 × 8, 8 × 16, 8 × 8, 8 × 4, 4 × 8, and 4 × 4 units. In particular, motion prediction performed with submacroblocks of 8x8 block size or less enables higher coding gain in complex images.

However, these seven variable block size motion prediction schemes of H.264 / AVC have a problem of increasing complexity compared to previous standards. In addition, the calculation amount is further increased by using multiple reference frames for motion prediction. However, despite the increase in the amount of calculation, the decrease in the prediction error is rather related to the characteristics of the image. Therefore, in some cases, the prediction gain that can be obtained by using more reference frames may be large, but otherwise, only a computation amount is wasted without obtaining a satisfactory gain.

Therefore, a number of methods for appropriately selecting the number of reference frames used for motion estimation while maintaining image quality have been proposed.

For example, a method of early termination through analysis of multiple reference frame gains, a method of using reference frames of adjacent blocks as candidate reference frames, a Kue's algorithm based on a reference frame search of 8 × 8 blocks, and an "Efficient reference frame selector for H.264 "IEEE Trans. Circuit Syst. Video Technol., Vol. 18, no 3, pp. 400-405, Mar. 2008).

However, such a conventional algorithm also has a problem that a lot of complexity is required to find an optimal reference frame.

Accordingly, an object of the present invention is to provide a method for appropriately selecting the number of reference frames used for mode estimation based on the correlation of the temporal complexity of the current frame and the previous frame in the multiple reference frame selection method for video encoding. Another object of the present invention is to provide a method of properly limiting the number of reference frames used for mode estimation by tracking a change in complexity of an encoding mode of a macroblock.

According to an aspect of the present invention, there is provided a method of selecting a multi-reference frame for video encoding, comprising: (a) performing motion estimation on a P16 × 16 block size of a current macroblock to be encoded with respect to a previous frame; calculating temporal complexity from the optimal reference frame information of the selected reference region through the motion estimation; And (c) adaptively selecting a candidate reference frame of the current macroblock according to the temporal complexity. As such, by appropriately selecting the number of candidate reference frames of the current macroblock according to the temporal complexity of the reference region of the previous frame, high-speed video encoding can be achieved while maintaining image quality.

In addition, the multiple reference frame selection method for video encoding, (d) determining the mode of the current macroblock; (e) comparing the complexity of the encoding mode of the reference region with the complexity of the encoding mode of the current macroblock; resetting optimal reference frame information of the current macroblock when the complexity of the encoding mode of the current macroblock increases; And (g) storing optimal reference frame information and mode information of the current macroblock, thereby preventing an error due to a change in the complexity of an encoding mode.

In the step (d), when the encoding mode is classified into CM_0, CM_1, CM_2, and CM_3, and the CM_0 consists of only SKIP mode, the CM_1 consists of SKIP mode and P16 × 16 mode, and the CM_2 corresponds to P8. When the block mode is configured with a block size larger than 8 blocks, the CM_3 includes a block mode smaller than the P8 block size. In step (e), the complexity of the encoding mode is CM_0, CM_1, CM_2, and CM_3. It is defined as increasing in ascending order, of course, can be classified and defined by other criteria.

In addition, step (d) is a step of determining a mode based on the complexity of the encoding mode. When the encoding mode of the reference region is the CM_0 or the CM_1, the RD (Rate Distortion) cost of a P16 × 16 block is determined. Performing a mode selection process of the remaining block modes using up to frames where the RD cost is no longer reduced; And when the coding mode of the reference region is the CM_2 or the CM_3, P8 × 4, P4 × 8, and P4 × are obtained by calculating the RD cost of the P8 × 8 block up to a frame where the RD cost no longer decreases. Performing a mode selection process of 4 blocks, and the P16 × 16, P16 × 8, and P8 × 16 blocks include performing a mode selection process on a reference frame selected from the P8 × 8 blocks, thereby again providing a range of candidate reference frames. By limiting this, faster encoding is possible.

In addition, the temporal complexity of the reference region is classified into Region_0, Region_1, Region_2, and Region_3, and in Region_0, when the optimal reference frame of the reference region consists only of ref0, the Region_1 indicates that the optimal reference frame of the reference region is ref0, In the case of ref1, the Region_2 represents a case where the optimal reference frame of the reference region consists of ref0, ref1, ref2, and the region_3 represents a case where the optimal reference frame of the reference region consists of ref0, ref1, ref2, ref3, ref4. . Here, ref0, ref1, ref2, ref3, and ref4 are given so that the index number increases as the distance from the previous frame increases.

 Meanwhile, according to the present invention, in the multi-reference frame selection method for video encoding, (a) obtaining a reference region by performing a motion estimation on the P16 × 16 block size of the current macroblock in the previous frames ; (b) adaptively selecting a candidate reference frame of the current macroblock corresponding to the temporal complexity of the reference region; And (c) comparing the encoding mode complexity of the reference region with the encoding mode complexity of the current macroblock to correct optimal reference frame information of the current macroblock when the encoding mode complexity of the current macroblock increases. It can be achieved by a reference frame selection method for video encoding, characterized in that.

Here, in the multi-reference frame selection method for video encoding, (d) determining a mode of the current macroblock, the range of the candidate reference frame for each mode based on the coding mode complexity of the reference region. It further includes the step of limiting, wherein step (d) is performed between step (b) and step (c).

As described above, according to the present invention, the complexity of motion estimation can be greatly reduced by appropriately selecting the number of reference frames used for mode estimation based on the correlation of the temporal complexity of the current frame and the previous frame. In addition, the complexity can be reduced without affecting the picture quality by appropriately limiting the number of reference frames used for mode estimation by tracking changes in the complexity of the encoding mode of the macroblock.

Hereinafter, specific embodiments of the present invention will be described with reference to the drawings.

Correlation of Temporal Complexity

The present invention focuses on the correlation of the temporal complexity between the selected reference region and the current macroblock through motion estimation of previous frames in order to obtain useful information needed for fast selection of the reference frame without additional computation.

3 illustrates a reference region of a previous frame selected by the motion vector of the current macroblock.

Referring to FIG. 3, the area where the car is located is far from the optimal reference frame and the current macroblock, which means that the current complexity of the macroblock is high. On the other hand, in the background part of FIG. 3, since only one previous frame is used as an optimal reference frame, it can be seen that the temporal complexity of the current macroblock is low.

From this fact, it can be seen that the temporal complexity of the reference region of the previous frame used for the motion estimation of the current macroblock establishes a high correlation between the temporal complexity of the current macroblock.

 Therefore, if the temporal complexity of the reference region is low, it may be assumed that the temporal complexity of the current macroblock is also low, and if the temporal complexity of the reference region is high, the current macroblock is encoded very temporally.

In order to prove this assumption, the reference areas are classified into four types. In the expression of the reference frame, ref0 means the frame closest to the coded frame, that is, the previous frame. As the number increases to ref1, ref2, ref3, etc., it means a frame that is far from the current frame.

FIG. 4 illustrates an example of classifying reference areas in a previous frame of the current macro block into four types.

Referring to FIG. 4, Region_0 is when all blocks of the reference region are ref0, Region_1 is when the reference region consists of ref0 and ref1, Region_2 is when the reference region consists of ref0, ref1, ref2, and Region_3 is ref0 Refers to the case consisting of, ref1, ref2, ref3, ref4.

Hereinafter, the statistics whether the temporal complexity between the current macroblock and the reference region are related to each other will be discussed. Table 1 shows the optimal reference frames of the current macroblock corresponding to each reference region (average of AKIYO, BUS, COASTGUARD, FORMAN, MOBILE, PARIS, and STEFAN pictures).

In Table 1 above, it can be seen that a high correlation is established between the Temporal Complexity of the Current Macroblock (TCCMB) and the Temporal Complexity of the Reference Region (TCRR). When the reference region is selected as Region_0, the ratio of ref0 selected as the optimal reference frame of the current macroblock is 90% or more. In Region_1, the ratio of ref0 is above 81% and the ratio of ref1 is above 9% on average. On the other hand, in Region_2 and Region_3, the ratio of ref2, ref3, and ref4 increases.

Therefore, the temporal complexity of the current macroblock can be predicted using the temporal complexity information of the reference region. For example, if the reference region of the current macroblock is Region_0, ref0 is used as a candidate reference frame of the current macroblock. If the reference region is Region_1, ref0 and ref1 are used as candidate reference frames. In the same way, if the reference region is Region_2, ref0, ref1, ref2 may be used as the candidate reference frame, and if the reference region is Region_3, ref0, ref1, ref2, ref3, ref4 may be used as the candidate reference frame.

In the above example, the temporal complexity is evaluated (calculated) in four ways, but in application, it can be evaluated in smaller or more steps, and the specific criteria can be changed.

Coding of Reference Area mode  Information

5 illustrates an example of a change in temporal complexity of the reference region in successive frames.

The change in the encoding mode in successive frames means that the characteristics of the object associated with the macroblock have changed. In this case, the correlation between the temporal complexity of the current macroblock and the temporal complexity of the reference region will also be changed. If so, the performance of the candidate reference frame selection method based on the correlation between the current macroblock and the temporal complexity of the reference region may be degraded.

In the present invention, in order to prevent such deterioration of performance, attention is paid to the complexity of the encoding mode and the change thereof. 6 illustrates an example of various reference regions for detecting a change in characteristics of a macroblock.

Referring to FIG. 6, the reference area is divided into four types according to the selected mode. Here, CM_0 refers to a case in which all macroblocks of the reference region are composed of only macroblocks of the SKIP mode, and CM_1 refers to a case in which macroblocks of the reference region are composed of macroblocks of the SKIP mode and the P16 × 16 mode. CM_2 refers to a case in which the block size of all macroblocks in the reference region is greater than or equal to the block size of the P8x8 mode, and CM_3 refers to a case in which the reference region is composed of macroblocks having a smaller block size than the P8x8 mode. Here, coding mode complexity (CMC) of CM_0, CM_1, CM_2, and CM_3 may be defined in ascending order. That is, the complexity of CM_0, CM_1, CM_2, and CM_3 increases as the number assigned to them increases.

Therefore, if the encoding mode complexity increases in successive frames, that is, if the encoding mode complexity of the current macroblock is increased compared to the encoding mode complexity of the reference region, the characteristics of the object may be changed. The performance of the candidate reference frame selection method based on the correlation between the temporal complexity of the reference region may be degraded.

Table 2 below shows data showing how there is a change in the correlation between the current macroblock and the temporal complexity of the reference region when the encoding mode complexity changes.

As can be seen from the above table, as the coding mode complexity increases in successive frames, the correlation between the current macroblock and the temporal complexity of the reference region is reduced. Therefore, it can be seen that the performance of the candidate reference frame selection method based on the correlation between the current macroblock and the temporal complexity of the reference region may be degraded. This error also affects later frames. The present invention further proposes a method for preventing the influence of such an error.

For example, if the encoding mode complexity of the reference region is increased, the optimal reference frame information of the current macroblock is reset, and the value of the maximum reference frame is stored for the next frame. For example, if the number of reference frames used in the encoder is five, the reference frames used in the corresponding region are reset to ref4, the maximum value. By doing so, it is possible to prevent an error that may occur when the current macroblock is used as a reference frame in a subsequent frame.

In the above example, the coding mode complexity is evaluated (calculated) in four ways, but in application, it can be evaluated in smaller or more steps, and the specific criteria can of course be changed.

RD ( Rate Distortion ) Coast  Early termination by change

When the candidate reference frame is determined through the above-described process, the mode determination of the current macroblock is performed with respect to the candidate reference frame. Since the present invention adaptively limits the number of candidate reference frames compared to the conventional method, the time taken for encoding is reduced accordingly.

Furthermore, the present invention proposes a method in which the current macroblock and the reference region terminate the mode decision early based on the association between the coding mode complexity.

The ratio of the encoding mode of the current macroblock varies according to the encoding mode complexity of the reference region. Table 3 below shows the encoding mode ratio of the current macroblock according to the encoding mode complexity of the reference region (AKIYO, BUS, COASTGUARD, FORMAN, MOBILE, PARIS, and STEFAN images).

Here, a = {SKIP, P16 × 16}, b = {P16 × 8 or P8 × 16}, and c = {P8 × 8, P8 × 4, P4 × 8, P4 × 4}.

Looking at the above table, it can be seen that the ratio of the current macroblock to c is larger when the reference region is CM_2 or CM_3 than when CM_0 or CM_1. From this fact it can be seen that in CM_2 and CM_3 it is advantageous to consider a reference frame selection method which focuses on small block size modes such as, for example, P8 × 8, P8 × 4, P4 × 8, and P4 × 4 modes. .

Let's look at the pattern of change in RD cost of item c of the current macroblock when the reference area is CM_2 or CM_3. If the RD cost of the current frame does not decrease compared to the RD cost of the previous frame, the RD cost value calculation is stopped in the remaining frame. Therefore, candidate reference frames used for motion estimation are limited to the current frame.

Table 4 below shows the accuracy (hit ratio) of the reference frame limited by the change pattern of the RD cost.

In Table 4, when the mode decision process is performed up to the reference frame in which the RD cost of P8x8 does not decrease, the hit ratio of the reference frame selection is significantly increased. Therefore, candidate reference frames in P8x8, P8x4, P4x8, and P4x4 modes are limited to reference frames in which the RD cost of P8x8 does not decrease. In the P16x16, P16x8, and P8x16 modes, candidate reference frames used in each P8x8 mode are used.

Example

The content of the present invention described above will be described through specific embodiments.

7 is a flowchart of an adaptive reference frame selection method according to a first embodiment of the present invention.

First, a reference region for a current macroblock is searched for in previous frames (S10). Since this search process is necessary for motion estimation, this does not increase the complexity. When the reference region is determined, a temporal complexity is calculated based on the optimal reference frame information of the reference region (S11). The farther the optimal reference frame of the reference region is, the greater the complexity is. The closer the frame is, the lower the complexity is.

When the temporal complexity of the reference region is determined as described above, the range of candidate reference frames of the current macroblock is adaptively determined according to the temporal complexity (S12). As described above, as the temporal complexity increases, the number of candidate reference frames of the current macroblock increases. On the contrary, as the temporal complexity decreases, the number of candidate reference frames of the current macroblock decreases.

When the range of the candidate reference frame is determined, a mode determination process of the current macroblock is performed within the range (S13). Since the number of candidate reference frames is reduced compared to the related art, the time required for mode determination will also be reduced.

However, in the present invention, in order to prevent an error, an opportunity to correct the reference frame information is provided. Referring to FIG. 7, the complexity of the encoding mode of the reference region and the complexity of the encoding mode of the current macroblock are evaluated (S14). If the complexity is increased, it is assumed that there is a change in the object, and the optimal reference frame information of the current macroblock is determined. Reset (S15, S16). As a result, an error may be prevented from occurring in a process of determining a reference frame performed in a subsequent frame.

Finally, the optimal reference frame information and mode information of the current macroblock are stored (S17), so that they can be used in subsequent frames.

8 is a flowchart of an adaptive reference frame selection method according to a second embodiment of the present invention.

First, the reference region of the current macroblock is searched for in the previous frame (S20). When a reference region is determined in the previous frame, it is classified into one of four types according to temporal complexity based on the optimal reference frame information of the reference region. If all blocks of the reference region have ref0 as an optimal reference frame, they are classified as Region_0 (S21), and only the immediately preceding frame is enabled for the current macroblock and used as candidate reference frames (S22).

If all blocks of the reference region have ref0 or ref1 as an optimal reference frame, they are classified as Region_1 (S23), and two frames of ref0 and ref1 for the current macroblock are enabled and used as candidate reference frames (S24). ).

On the other hand, if all blocks of the reference region have ref0, ref1, and ref2 as optimal reference frames, they are classified as Region_2 (S25), and a candidate reference is performed by enabling three frames of ref0, ref1, and ref2 for the current macroblock. It is used as a frame (S26).

Finally, if the block of the reference region has ref0, ref1, ref2, and ref3 or ref4 as the best reference frame, it is classified as Region_3, and four frames of ref0, ref1, ref2, ref3 or ref4 for the current macroblock are classified. Enable it and use it as a candidate reference frame (S27).

In the above description of the selection of candidate reference frames, it should be noted that there is a difference between the index of the optimal reference frame of the reference region (relative to the previous frame) and the index of the reference frame used in the current frame (relative to the current frame). . For example, even if it is ref0, ref0 of the previous frame reference and ref0 of the current frame reference are different.

Next, the determined candidate reference frames are searched to determine an optimal encoding mode of the current macroblock. In the present invention, an optimal mode determination process is performed according to the encoding mode complexity (S28). For example, when the mode selected in the reference region is CM_0 or CM1, the RD cost of the P16 × 16 block is obtained, and if the RD cost is no longer reduced, the mode selection process of the remaining modes is performed using the frame. On the other hand, if the mode selected in the reference region is CM_2 or CM3, the RD cost of the P8 × 8 block is obtained, respectively. The mode selection process is performed, and the P16x16, P16x8, and P8x16 blocks perform the mode selection process in the reference frame used in the P8x8 block. As a result, the range of candidate reference frames is once again limited to perform mode determination more quickly.

In this way, when the optimal reference frame and mode of the current macroblock are determined, the complexity of the encoding mode of the reference region is compared with the complexity of the mode determined for the current macroblock, and if the encoding mode complexity is increased in the reference region of the previous frame, The temporal complexity of the macroblock, that is, the optimal reference frame information is reset (S29 and S30) so that an error that may occur when the next frame is encoded is not affected.

Information about the optimal reference frame and mode of the current macroblock thus determined or reset is stored for use as the encoding mode complexity in the next frame (S31).

Experiment result

H.264 / AVC J.M 12.4 was used to investigate the performance of the fast reference frame selection method according to the present invention. In the experiment, six video sequences with small to large motion, and AKIYO, BUS, COASTGUARD, FORMAN, MOBILE, and PARIS sequences with CIF size 352 × 288 were used. Seven interblock modes were used and three quantization parameters (QP = 24, 28, 32) were tested. In addition, all experiments were performed on the Intel Pentium Core 2 Duo 3.16 GHz with 3.25 GB of RAM. Other coding parameters are shown in Table 5 below.

ΔPSNR (PSNR Difference) and ΔBitrate (Bitrate Difference) were used to compare the performance. The following equation was used to calculate ΔTime (Time Difference) or ΔME Time (Motion Estimation Difference).

Where Time _{Ref .} Means the encoding time used by the reference method, Time _Prop. Represents the time spent by the method of the present invention.

In order to compare the RD and speed performance according to the present invention, a fast full search (FFS) technique by JM reference software was used as the reference method. Table 6 below is a result of comparing the coding performance according to the present invention with the conventional Kue's algorithm.

Looking at the above tables, it can be seen that the method according to the invention achieves high speed performance with minimal loss in image quality and a negligible bit rise. At QP = 24, Kue's algorithm averaged 42% less time for coding than the reference method, with PSNR loss of 0.048dB on average and 2.29% bitrate. On the other hand, the method according to the present invention has reduced the encoding time by an average of 53%, an average loss of 0.045 dB in the PSNR, and an average 2.35% increase in the bit rate.

At QP = 28, the Kue's algorithm reduced the average time to coding by 42%, lost PSNR by 0.06dB, and increased the bitrate by 2.13%. In the method according to the present invention, the time taken for coding is reduced by 55% on average, there is an average loss of 0.006 dB in PSNR, and the bit rate is increased by 2.33% on average.

At QP = 32, the Kue's algorithm reduced the total time to coding by an average of 43%, the PSNR lost an average of 0.09dB, and the bitrate increased by 1.24%. In the method according to the present invention, the total time taken for coding was reduced by an average of 58%, an average loss of 0.09 dB in the PSNR, and an increase in the bit rate by 1.85%.

In order to accurately analyze the complexity, the complexity of the frames used in each method was compared. Specifically, the complexity of the SAD calculation for 4 × 4 blocks was compared with that of the FFS method. The performance comparison between the frame used and the SAD calculation is shown in Table 7 below.

As can be seen from the table above, the method according to the present invention outperforms Kue's algorithm in terms of the number of frames used and the calculated SAD.

9 to 12 are graphs comparing RD performance with frames used in Foreman and Stefan sequences. Referring to these figures, it can be seen that the method according to the invention is similar to Kue's in RD performance. However, in terms of speed, the present invention far exceeds Kue's algorithm.

As described above, the reference frame selection method based on the information of the reference region used for the motion estimation of the previous frame according to the present invention shows a good performance because the complexity is greatly reduced while there is almost no RD loss. have.

The present invention described above may be implemented by a software algorithm, and may also be implemented by a device including a memory in which the algorithm is stored, a processor on which the algorithm is executed, and the like. It is also applicable to all displays which process and display an image.

In particular, the present invention can be applied to all applications using the video compression technology, for example, a mobile phone supporting a mobile broadcast and many DMB players, cameras, HDTV, IPTV, PC, such as many devices that require a video compression technology.

Although some embodiments of the invention have been shown and described, it will be apparent to those skilled in the art that modifications may be made to the embodiment without departing from the spirit or spirit of the invention. . It is intended that the scope of the invention be defined by the claims appended hereto and their equivalents.

1 schematically illustrates the use of multiple reference frames in H.264.

2 is a diagram for explaining a motion prediction method of seven variable block sizes of H.264.

3 illustrates a reference region of a previous frame selected by the motion vector of the current macroblock.

4 illustrates an example of dividing the reference region in the previous frame of the current macroblock into four types.

5 illustrates an example of a change in temporal complexity of the reference region in successive frames.

6 illustrates an example of various reference regions for detecting a change in characteristics of a macroblock.

7 is a flowchart of an adaptive reference frame selection method according to a first embodiment of the present invention.

8 is a flowchart of an adaptive reference frame selection method according to a second embodiment of the present invention.

9 to 12 are graphs comparing RD performance with frames used in Foreman and Stefan sequences.

Claims (7)

In the multiple reference frame selection method for video encoding, (a) performing motion estimation on a P16 × 16 block size of a current macroblock to be encoded to at least one of previous frames; calculating temporal complexity from the optimal reference frame information of the selected reference region through the motion estimation; And (c) adaptively selecting candidate reference frames of the current macroblock according to the temporal complexity. The method of claim 1, The multi-reference frame selection method for video encoding, after step (c) (d) determining a mode of the current macroblock; (e) comparing the complexity of the encoding mode of the reference region with the complexity of the encoding mode of the current macroblock; resetting optimal reference frame information of the current macroblock when the complexity of the encoding mode of the current macroblock increases; And and (g) storing optimal reference frame information and mode information of the current macroblock. The method of claim 2, In step (d), the encoding mode is classified into CM_0, CM_1, CM_2, and CM_3, and CM_0 consists of only SKIP mode, CM_1 consists of SKIP mode and P16 × 16 mode, and CM_2 is P8 × 8. When configured with a block mode of block size or more, the CM_3 is a case that includes a block mode of less than the P8 × 8 block size, In step (e), the complexity of the encoding mode is defined as increasing in ascending order of CM_0, CM_1, CM_2, and CM_3. The method of claim 3, Step (d) is a step of determining a mode based on the complexity of the encoding mode, When the coding mode of the reference region is the CM_0 or the CM_1, the RD (Rate Distortion) cost of the P16 × 16 block is obtained, and the mode selection process of the remaining block modes is performed using the frame up to which the RD cost is no longer reduced. Performing; And If the coding mode of the reference region is the CM_2 or the CM_3, P8 × 4, P4 × 8, and P4 × 4 are obtained by calculating the RD cost of the P8 × 8 block to the frame where the RD cost no longer decreases. Performing a mode selection process of the block, and the P16 × 16, P16 × 8, and P8 × 16 blocks include performing a mode selection process on a reference frame selected from the P8 × 8 block. How to select a reference frame. The method according to any one of claims 1 to 4, The temporal complexity of the reference region is classified into Region_0, Region_1, Region_2, and Region_3, and in Region_0, if the optimal reference frame of the reference region consists only of ref0, Region_1 indicates that the optimal reference frame of the reference region is ref0, ref1. In this case, the Region_2 indicates that the optimal reference frame of the reference region is composed of ref0, ref1, ref2, and the Region_3 represents a case where the optimal reference frame of the reference region is composed of ref0, ref1, ref2, ref3, ref4. Multiple reference frame selection method for video encoding. (Here, ref0, ref1, ref2, ref3, ref4 are given so that the index number increases as the distance from the previous frame increases) In the multiple reference frame selection method for video encoding, (a) obtaining a reference region by performing motion estimation on a P16 × 16 block size of a current macroblock in previous frames; (b) adaptively selecting a candidate reference frame of the current macroblock corresponding to the temporal complexity of the reference region; And (c) comparing the encoding mode complexity of the reference region with the encoding mode complexity of the current macroblock and correcting optimal reference frame information of the current macroblock when the encoding mode complexity of the current macroblock increases. Multiple reference frame selection method for video encoding characterized in that. The method of claim 6, In the multi-reference frame selection method for video encoding, (d) determining a mode of the current macroblock, and once again the range of the candidate reference frame for each mode based on the encoding mode complexity of the reference region. The method further comprises the step of limiting, wherein step (d) is performed between step (b) and step (c).

Images