CN104581152A - HEVC intra-frame prediction mode decision accelerating method - Google Patents

Abstract

An HEVC intra-frame prediction mode selection acceleration method, first guessing the texture consistency between the spatially adjacent PU and the current PU, if there is texture consistency, for the prediction unit of the first mode selected by RMD in the MPM, skip In the RDO process, directly set this mode as the RDO optimal mode. Secondly, using the texture features in the PU reflected by the first two modes selected by RMD, by judging whether the first two modes selected by RMD are both DC or Planar modes, or both are corner modes, the first mode selected by RMD is not in MPM The prediction unit in is further divided into different cases for mode selection acceleration. The invention fully exploits the texture consistency of the prediction unit reflected by RMD and MPM and the adjacent coded prediction unit, as well as the texture characteristics in the prediction unit, and has the advantages of effectively reducing the complexity of the RDO prediction mode selection process in intra-frame mode selection. .

Description

A HEVC Intra Prediction Mode Selection Acceleration Method

technical field

The present invention relates to the field of high-performance video coding (High Efficiency Video Coding, HEVC), in particular to the technology of intra-frame mode selection in coding.

Background technique

Digital video has become an integral part of our daily lives. With the rapid development of network and video capture technology, ultra-high-definition video service has become the development trend of digital video in the future. In order to better meet the needs of ultra-high-definition video services, the Moving Picture Experts Group (MPEG) and the Joint Collaborative Team on Video Coding (JCT-VC) jointly launched a new video compression standard—— High Efficiency Video Coding (HEVC). As the successor of ITU-T H.264/MPEG-4AVC, HEVC adopts many new coding techniques, which can save half of the codewords compared with the predecessor with the same video quality. Due to the adoption of many new technologies, the complexity of the encoder is greatly increased, which makes the realization of real-time encoding more and more difficult, so reducing the complexity of the encoder is an important research field of HEVC.

In order to improve the compression performance of intra-frame coding, HEVC increases the number of intra-frame prediction directions to 35, 0 and 1 modes are DC and Planar modes, and the remaining 2-35 modes are corner modes, which greatly increases the number of intra-frame prediction directions. Computational complexity of choice. In order to reduce the complexity of this part, the existing technology adopted by HEVC is: for a prediction unit (PredictionUnit, PU), the first thing is a simple rough mode selection (Rough Mode Decision, RMD). RMD uses Hadamard Transform to calculate the rate-distortion cost of each mode. RMD arranges the 35 modes according to the rate-distortion cost from small to large. Among them, the encoder will select the RDO alternative mode according to the size of the PU. For PU units of 64x64, 32x32, and 16x16 sizes, RMD selects the three modes with the lowest cost as RDO alternative modes. For PU units of 8x8 and 4x4 sizes , RMD selects 8 modes with the lowest cost as RDO candidate modes, and these candidate modes will be stored in the candidate list. After the RMD ends, the Most Probable Mode (Most Probable Mode, MPM) will check whether the optimal mode selected by the PU above and to the left of the current PU is in the list of alternative modes, if the optimal mode of the PU above or to the left is If it is not in the list of alternative modes, add it to the list. For details, please refer to Document 1 (see L.Zhao, L.Zhang, S.Ma, and D.Zhao, "Fast mode decision algorithm for intra prediction in HEVC, "in Proc. in Proc. VCIP 2011, pp.1-4.). In the next rate-distortion optimization (Rate-distortion optimization, RDO) stage, the discrete cosine/sine transform (DCT/DST) used for encoding will be used to calculate the rate-distortion cost for all alternative modes. The mode with the smallest rate-distortion cost in the RDO stage is the final optimal mode. After determining the optimal mode, the Residual Quadtree transform (RQT) will transform and encode the PU according to the selected optimal mode. In addition to HEVC's existing fast intra-frame mode selection technology, there are also some other methods. For example, literature 2 (see M. Zhang, C. Zhao, and J. Xu, "An adaptive fast intra mode decision in HEVC," in Proc. ICIP 2012, pp. 221-224.) distribution of alternative modes selected by RMD to infer the texture features of the current prediction block. This technology requires two different processing schemes for PUs of 16x16 and 32x32 sizes and PUs of 8x8 and 4x4 sizes, and for PUs of 8x8 and 4x4 sizes, it is necessary to analyze the first four alternative modes selected by RMD and pass judgment Whether there is a DC or Planar mode, and whether the first candidate corner mode and the second candidate corner mode are adjacent to infer the texture characteristics of the PU, and this technique does not use the relationship between the candidate mode selected by RMD and the MPM Reflected texture consistency of adjacent PUs.

Contents of the invention

The purpose of the present invention is to provide an HEVC intra prediction mode selection acceleration method.

To achieve the above object, the solution of the present invention is: firstly, focus on the texture consistency between the spatially adjacent PU and the current PU. Using the feature of texture consistency, for the prediction unit of the first mode selected by RMD in the MPM, the RDO process is skipped, and the mode is directly set as the RDO optimal mode. Secondly, using the texture features in the PU reflected by the first two modes selected by RMD, by judging whether the first two modes selected by RMD are both DC or Planar modes, or both are corner modes, the first mode selected by RMD The prediction units that are not in the MPM are further divided into flat blocks, texture blocks with a certain direction, and complex texture blocks without a clear direction for mode selection acceleration.

The invention fully exploits the texture consistency of the prediction unit reflected by the RMD and the MPM and the adjacent coded prediction unit and the texture characteristics in the prediction unit, and can effectively reduce the complexity of the RDO prediction mode selection process in the intra-frame mode selection.

The present invention uses the texture features of the adjacent PU of the current prediction unit reflected by the MPM, and the texture feature of the current PU reflected by the first candidate mode selected by the RMD, and combines the MPM and RMD to estimate the current PU and the adjacent PU. Whether the texture is consistent. The criterion of consistency is defined as: if the first candidate mode is in the MPM, then the current PU is considered to be consistent with the texture of the adjacent PU. For a prediction unit that has texture consistency with an adjacent PU, the first candidate mode of RMD will be the optimal mode of RDO for the current prediction unit.

For the prediction unit that is judged to be inconsistent with the adjacent PU texture according to the above consistency criterion, the present invention will use the first two candidate modes selected by RMD to infer the current PU texture feature, and divide the prediction unit into three categories according to the following criteria :

(1) If the first and second candidate modes obtained by the current prediction unit RMD are both DC or Planar modes, then the PU is a flat block;

(2) If the first and second candidate modes obtained by the current prediction unit RMD are both corner modes, then the PU is a texture block with a certain direction;

(3) If the first and second candidate modes obtained by the current prediction unit RMD are not both DC or Planar modes, nor are both corner modes, then the PU is a complex texture block without a clear direction.

For the first type of prediction unit, its RDO candidate modes only include DC and Planar modes, and other modes can be removed from the candidate list. For the second type of prediction unit, its RDO candidate modes only include the first three candidate modes and MPM obtained by RMD, and other modes can be removed from the candidate list. For the third type of prediction unit, its RDO candidate modes only include the first two candidate modes and MPM obtained by RMD, and other modes can be removed from the candidate list. For this reason, the implementation steps of the technical solution of the present invention are:

Step 1: Perform RMD and MPM of the current prediction unit.

Step 2: Check whether the first candidate mode selected by the RMD is in the MPM. If the first candidate mode selected by RMD is in the MPM, then the predicted and adjacent coded units have similar texture features, and this mode is directly set as the RDO optimal mode, skipping the RDO process.

Step 3: For the prediction unit whose first candidate mode selected by the RMD is not in the MPM, check whether the first two candidate modes selected by the RMD are both DC or Planar modes. If the first two candidate modes selected by RMD are both DC or Planar modes, the prediction is a flat block, and its RDO only checks the two candidate modes of DC and Planar, and other corner modes can be removed from the list of candidate modes.

Step 3: For the first candidate mode selected by RMD is not in the MPM, and the first two candidate modes selected by RMD are not both prediction units of DC or Planar mode, check the first two candidate modes selected by RMD Is it all angular mode. If the first two candidate modes selected by RMD are both angle modes, then the prediction unit is a texture block with a certain direction, RDO will only check the first three candidate modes and MPM selected by RMD, other modes can be obtained from the removed from the selection mode list.

Step 4: For the first candidate mode selected by RMD is not in the MPM, and the first two candidate modes selected by RMD are not both DC or Planar mode, nor are they all prediction units of angular mode, then the prediction unit It is a complex texture block without a clear direction. RDO will only check the first two candidate modes and MPM selected by RMD, and other modes can be removed from the list of candidate modes.

Step 5: Perform RQT on the current coding unit according to the optimal mode selected by RDO.

Owing to adopting said scheme, the beneficial effect of the present invention is:

1. This discovery first utilizes the relationship between the texture consistency of the adjacent prediction blocks reflected by the RMD and MPM results and the current prediction block, so that the texture consistency prediction unit can directly skip the RDO process. This technology can effectively reduce the complexity of HEVC intra mode selection process.

2. This algorithm adopts a unified intra-block texture characteristic prediction scheme for prediction units of all sizes, and only needs to analyze the first two RDO candidate modes obtained by RMD to divide the prediction units into three categories according to different texture characteristics, and perform Forecast mode select Accelerated.

3. On the basis of the existing HEVC intra-frame mode selection scheme, the present invention can effectively reduce the encoding complexity without losing the subjective and objective quality of the video, and the method is easy to implement.

Description of drawings

FIG. 1 is a flow chart of the HEVC intra-frame fast mode selection method of the present invention.

Detailed ways

The present invention will be further described below in conjunction with the embodiments shown in the accompanying drawings.

As shown in Figure 1, an embodiment of the present invention includes the following steps:

Step 1: Based on the HEVC common test platform HM, after the RMD and MPM of the current prediction unit are completed, check the current coding block according to the list of alternative modes obtained by RMD and the most probable mode of adjacent coding blocks obtained by MPM Texture relationship with adjacent coded blocks. If the first mode in the candidate mode list is in the MPM, the RDO is skipped, and the first candidate mode is used as the optimal mode of the RDO, and the subsequent RQT process is directly performed. Otherwise go to step 2. This step corresponds to the first branch in Figure 1, that is, the first candidate mode is in the most probable mode.

Step 2: Check whether the first two modes in the list of alternative modes are both 0, 1 modes, namely DC and Planar modes. If so, RDO only checks the 0,1 mode, and after obtaining the optimal prediction mode, the subsequent RQT process is performed. Otherwise go to step 3. This step corresponds to the second branch in Figure 1, that is, the first and second alternative modes are modes 0 and 1.

Step 3: Check whether the first two modes in the candidate mode list are both angle modes, that is, whether they are both within the closed interval of 2-35. If so, RDO only checks the first three modes and MPM in the candidate list, and after obtaining the optimal prediction mode, the subsequent RQT process is performed. Otherwise go to step 4. This step corresponds to the third branch in Figure 1, that is, the first three candidate modes plus the most probable mode.

Step 4: RDO only checks the first two modes and MPM in the candidate list, and after obtaining the optimal prediction mode, the subsequent RQT process is performed. This step corresponds to the fourth branch in Figure 1.

The above description of the embodiments is for those of ordinary skill in the art to understand and apply the present invention. It is obvious that those skilled in the art can easily make various modifications to these embodiments, and apply the general principles described here to other embodiments without creative efforts. Therefore, the present invention is not limited to the above-mentioned embodiments. Improvements and modifications made by those skilled in the art according to the disclosure of the present invention without departing from the scope of the present invention should fall within the protection scope of the present invention.

Claims (7)

1. A HEVC intra-frame prediction mode selection acceleration method, characterized in that: firstly, the texture consistency between the spatially adjacent PU and the current PU is estimated, if there is texture consistency, the first mode selected for RMD is in the MPM The prediction unit skips the RDO process and directly sets this mode as the RDO optimal mode. 2. The HEVC intra prediction mode selection acceleration method according to claim 1, characterized in that: the criterion of texture consistency is defined as: if the first candidate mode is in the MPM, then the current PU and the adjacent PU is texture consistent. 3. The HEVC intra prediction mode selection acceleration method according to claim 1, characterized in that: using the texture characteristics of the adjacent PU of the current prediction unit reflected by the MPM, and the first alternative mode selected by the RMD The texture features of the current PU, combining MPM and RMD to infer whether the texture of the current PU is consistent with that of adjacent PUs. 4. The HEVC intra-frame prediction mode selection acceleration method according to claim 1, characterized in that: the texture features in the PU reflected by the first two modes selected by RMD are used to determine whether the first two modes selected by RMD are both If it is DC or Planar mode, or both are angle modes, the prediction unit whose first mode selected by RMD is not in MPM is further divided into different cases to perform mode selection acceleration. 5. the HEVC intra prediction mode selection acceleration method according to claim 4, is characterized in that: For prediction units that are judged to be inconsistent with adjacent PU textures, the first two candidate modes selected by RMD are used to infer the current PU texture features, and the prediction units are divided into three categories according to the following criteria: (1) If the first and second candidate modes obtained by the current prediction unit RMD are both DC or Planar modes, then the PU is a flat block; (2) If the first and second candidate modes obtained by the current prediction unit RMD are both corner modes, then the PU is a texture block with a certain direction; (3) If the first and second candidate modes obtained by the current prediction unit RMD are not both DC or Planar modes, nor are both corner modes, then the PU is a complex texture block without a clear direction. 6. The HEVC intra prediction mode selection acceleration method according to claim 5, characterized in that: for the first type of flat block prediction unit, its RDO alternative modes only include DC and Planar modes, and other modes can be selected from the alternative Remove from the list; for the second type of texture block prediction unit with a certain direction, its RDO candidate modes only include the first three candidate modes and MPM obtained by RMD, and other modes can be removed from the candidate list; for the third For a complex texture block prediction unit with no clear direction, its RDO candidate modes only include the first two candidate modes and MPM obtained by RMD, and other modes can be removed from the candidate list. 7. The HEVC intra prediction mode selection acceleration method according to claim 1, characterized in that: comprising the following steps: Step 1: Perform RMD and MPM of the current prediction unit; Step 2: Check whether the first candidate mode selected by RMD is in the MPM; if the first candidate mode selected by RMD is in the MPM, then the prediction and the adjacent coded unit have similar texture features, and directly This mode is set to RDO optimal mode, skipping the RDO process; Step 3: For the prediction unit whose first candidate mode selected by RMD is not in the MPM, check whether the first two candidate modes selected by RMD are both DC or Planar modes; if the first two candidate modes selected by RMD Both are DC or Planar modes, then the prediction unit is a flat block, and its RDO only checks the two alternative modes of DC and Planar, and other corner modes can be removed from the list of alternative modes; Step 4: For the first candidate mode selected by RMD is not in the MPM, and the first two candidate modes selected by RMD are not both DC or Planar mode prediction units, check the first two candidate modes selected by RMD Whether they are all angle modes; if the first two candidate modes selected by RMD are all angle modes, then the prediction unit is a texture block with a certain direction, and RDO will only check the first three candidate modes and MPM selected by RMD , other patterns can be removed from the list of alternative patterns; Step 5: If the first candidate mode selected by RMD is not in the MPM, and the first two candidate modes selected by RMD are not both DC or Planar mode, nor are both prediction units of angle mode, then the prediction unit It is a complex texture block without a clear direction. RDO will only check the first two alternative modes and MPM selected by RMD, and other modes can be removed from the alternative mode list; Step 6: Perform RQT on the current coding unit according to the optimal mode selected by RDO.