CN104796694B - Optimization intraframe video coding method based on video texture information - Google Patents

Abstract

An optimized intra-frame video coding method based on video texture information in the field of video coding. Through the calculation and mapping of I-frame gradient information, the video image gradient information and its actual meaning are used to calculate all pixels of the intra-frame coding frame, and for high-efficiency video The intra prediction mode is used for mapping; the coding unit is divided, and the attribute of the coding unit is determined according to the gradient information of the pixels in the current coding unit, and the attribute and direction consistency of the four smaller coding units included in the larger coding unit are determined from bottom to top The division method skips unnecessary division traversal; the fast mode selection adopts the offline training method to predict and optimize the prediction mode selection under different coding unit attribute conditions. The invention optimizes the mode selection and coding unit division of the intra-frame video coding in combination with the texture characteristics of the video, improves the coding speed under the premise of ensuring the coding quality, and ensures the stability of the algorithm for coding different video sequences.

Description

Optimal Intra-Frame Video Coding Method Based on Video Texture Information

technical field

The present invention relates to an intra-frame video coding method in the technical field of video coding, in particular to an optimized intra-frame video coding method based on video texture information.

Background technique

Video coding technology is to compress and convert video into video code stream through certain technology according to compression standard. At present, for different video application fields and technology development, there are many video coding standards, mainly MPEG-x series and H.26x series, such as MPEG-2, MPEG-4 standards, H.263, H.264 standards, etc. , In addition, there is the audio and video coding standard AVS formulated by the domestic digital audio and video coding and decoding technology standard working group. With the development of multimedia, people's demand for 3D, high-definition, ultra-high-definition and other videos has increased. Followed by higher requirements for video coding efficiency, HEVC (high efficiency video coding, high-efficiency video coding) was born, which achieved a nearly 50% improvement in coding efficiency compared to H.264. The AVS working group has also formulated a new generation of high-efficiency video coding standard, AVS2. It has reached a video coding standard that is similar to HEVC, and even more efficient in some video scene applications. In pursuit of the improvement of coding quality, high-efficiency video coding introduces a large number of new coding technologies, which greatly increases the computational complexity of coding while improving the quality, especially the intra-frame coding module in video coding, which reduces the coding speed. Optimizing high-efficiency video coding standards is a research hotspot in video coding research.

Spatial correlation and temporal correlation are important basis for video compression. In intra coding, spatial correlation can often play an important role in the coding process. Video texture information is the embodiment of spatial correlation, and is widely used in image processing and video coding. According to the search and understanding of the prior art, methods such as edge information, spatial correlation statistics, and gradient information are all important applications of spatial correlation. In the video coding process, using video texture information to predict the coding method can reasonably reduce the prediction and segmentation modes that need to be traversed in the coding process. Under the premise of accurate prediction, the computational complexity of high-efficiency video coding is greatly reduced, so as to achieve high efficiency. Reasonable optimization of video encoders. The simplification and replacement of the encoding process will often lead to a decrease in the quality of video encoding, and unreasonable modification of the encoding framework will also cause the instability of the encoder. How to accurately apply video texture information and rationally combine with video coding to achieve accurate prediction and reduce the complexity of video coding is an important topic in current coding research.

After searching the prior art, it was found that Chinese Patent Document No. CN103517069A, public (announcement) date 2014.01.15, discloses a fast mode selection method for HEVC intra-frame prediction based on texture analysis. Before the intra-prediction, determine the main texture direction and texture complexity of each 4×4 unit in the coding tree unit according to the absolute value of the gradient in the four directions of horizontal, vertical, lower left, and lower right, and use a relatively smooth area according to the texture Large coding units and areas with complex textures use the principle of smaller coding units to determine the division of the current coding tree unit. When predicting, according to the main texture direction of the prediction unit, several least likely prediction modes are excluded, and then rough mode selection and rate-distortion optimization mode selection are performed according to the HEVC coding standard. The HEVC intra prediction fast mode selection method based on texture analysis proposed by this technology can significantly improve the encoding speed while ensuring the encoding quality. However, compared with the present invention, this prior art cannot solve technical problems including direct and accurate prediction direction estimation, simplified mode selection list, prediction mode selection adjustment and coding unit division for different video sequence feature processing. In terms of intra-frame coding stability for different video sequences, the present invention has stronger stability and accuracy.

Chinese Patent Document No. CN103096090A, Publication (Announcement) Day 2013.05.08, discloses a method for encoding block division in video compression, which is characterized in that it includes the following steps: read the pixel values in the entire LCU, and complete the block Merge the search table; enter the CU of each depth, and obtain the depth and position information of the combined block corresponding to the search table according to the depth and position information; if the current CU depth is consistent with the combined block depth, perform a bottom-up block division method Judgment process; otherwise, perform the judgment process of the block division fast algorithm from top to bottom. Using this technology combined with two aspects, the block division fast algorithm proposed greatly speeds up the encoding speed and improves the encoding efficiency under the premise of ensuring that the video quality and output bit rate of the HEVC encoder are basically unchanged. However, compared with the present invention, this prior art has unsolvable technical problems including fast and intuitive application of video spatial correlation and texture information. The search table method is computationally complex and has low optimization, and this prior art only corrects the depth division. The present invention uses a more concise and intuitive calculation method to combine prediction and depth division, and the optimization speed and performance are greatly improved.

Contents of the invention

Aiming at the above-mentioned deficiencies in the prior art, the present invention proposes an optimized intra-frame video coding method based on video texture information, which uses the gradient information of the video sequence to screen the prediction modes of intra-frame coding, reducing the need for a large number of complex Calculated rate-distortion decision candidates. A bottom-up partition optimization for the recursive coding unit partition framework in high-efficiency video coding is adopted. According to the video texture complexity, the coding unit is divided into smooth and rough two attributes, and then the combination and division of the coding unit are determined in a targeted manner. At the same time, in order to ensure the stability of the encoding framework for encoding efficiency of different video sequences, based on experiments, different numbers of prediction mode candidates are selected for encoding units with different attributes, thereby greatly improving the speed of video encoding while ensuring stable and excellent encoding quality. .

The present invention is achieved through the following technical solutions:

The present invention comprises the following steps:

The first step is to calculate the gradient of all pixels in the intra-frame coding frame, that is, the I frame, to obtain the gradient value representing the direction of pixel motion; by comparing the relationship between the intra-frame prediction direction map and the gradient, the gradient and prediction mode mapping relationship diagram is obtained, Thus, the optimal prediction mode of each pixel is obtained during the encoding process;

The second step is to classify the attributes of the coding blocks according to the video texture information, and then merge the coding units with the same attributes, and then divide the optimized coding units into adjustable coding units;

The attribute classification refers to: making statistics on the optimal prediction modes of each pixel in the current coding unit, and when there is a mode selection rate exceeding 80% in the statistical results, it is judged that in the current coding unit, the pixel has a uniform operation direction, the image texture is relatively simple, set the attribute of the coding unit as a smooth unit; otherwise, set the attribute of this unit as a rough unit.

The merging refers to: analyzing and judging from the smallest coding unit 8x8 module, when four consecutive coding units of the same size are all smooth and their main gradient directions are consistent, then these four units can be combined into one relatively A large unit, and the larger coding unit is also set as a smooth unit, and the gradient direction is consistent with it; otherwise, the size of the current coding unit is selected, and the merge operation is not performed on it.

The third step is to set different mode candidate values according to the different attributes of the coding unit for prediction, and then select the corresponding number of modes in the gradient direction calculated by the gradient mapping statistics according to the mode frequency from large to small for RDcost calculation, and select RDcost The smallest mode is the best predictive mode.

For the above prediction, firstly, through offline training, different sequences with different resolutions are selected for testing respectively, and the best balance point between encoding speed and quality is selected according to the encoding result, and then group comparison tests are carried out around the balance point, and the experimentally selected The number of best mode candidates that should be selected by the smooth unit and the rough unit respectively.

technical effect

Compared with the prior art, the present invention combines video texture information to optimize prediction mode selection and coding unit division of intra-frame coding, thereby reducing coding complexity and improving coding speed. The whole coding process is divided into two main parts, fast mode selection and fast coding unit division. Through the gradient information of the video texture, and its relationship with the prediction direction and the division of the coding unit, the calculation process of the prediction mode selection is simplified, and the division of the coding unit is predicted to avoid unnecessary coding prediction calculations, thereby improving transcoding Speed saves time.

Description of drawings

Fig. 1 is a schematic flow chart of the present invention.

Figure 2 is a comparison chart of the serial BQ Square performance curve;

In the picture: QP are 27, 32, 38, 45 respectively.

detailed description

The embodiments of the present invention are described in detail below. This embodiment is implemented on the premise of the technical solution of the present invention, and detailed implementation methods and specific operating procedures are provided, but the protection scope of the present invention is not limited to the following implementation example.

Example 1

As shown in Figure 1, this embodiment includes the following steps:

Step 1. Gradient calculation is performed on all pixels of the intra-frame coding frame, that is, the I frame, to obtain a gradient value representing the direction of motion of the pixel; by comparing the relationship between the intra-frame prediction direction map and the gradient, the gradient and the prediction mode mapping relationship diagram are obtained, thereby The optimal prediction mode for each pixel is obtained during the encoding process.

In order to ensure the accuracy of prediction, HEVC introduces a large number of intra-frame prediction directions. For example, the brightness module in HEVC uses 33 directional prediction modes and two special prediction modes (DC, planar), and AVS2 uses 30 directions. The directional prediction mode and three special prediction modes (DC, plane, bilinear), the above gradient value and the intra prediction directional mode in HEVC have a one-to-one mapping relationship.

Step 2: Analyze and judge the coding block according to the video texture information, so as to make a fast division method selection, specifically:

2.1) Make statistics on the optimal prediction mode of each pixel in the current coding unit, and when there is a mode selection rate exceeding 80% in the statistical results, it is judged that in the current coding unit, the pixels have a unified running direction, and the image texture is relatively simple , set the property of the coding unit as a smooth unit; otherwise, set the property of the unit as a rough unit.

2.2) Analyze and judge from the smallest coding unit 8x8 module. When four consecutive coding units of the same size are smooth and their main gradient directions are consistent, these four units can be combined into a larger unit, and the The larger coding unit is also set as a smoothing unit, and the gradient direction is consistent with it; otherwise, the size of the current coding unit is selected and not merged.

2.3) The optimized coding unit is divided into adjustable coding units; based on the recursive calculation method in HEVC, the main computational complexity starts from the innermost minimum coding unit, that is, the 8x8 coding unit, so the design in the present invention consists of The bottom-up division judgment mode can minimize the amount of calculation.

Step 3. Quick mode selection, the specific steps include:

3.1) Using offline training, select different sequences with different resolutions to test separately, select the best balance point between encoding speed and quality according to the encoding results, and then conduct group comparison tests around the balance point, and choose smooth units and rough units in the experiment. The number of best pattern candidates that the unit should choose respectively.

3.2) In the gradient direction calculated by the gradient mapping statistics, select the corresponding number of modes according to the mode occurrence frequency from large to small for RDcost (Rate Distortion, rate distortion cost) calculation, and select the mode with the smallest RDcost, which is the best prediction mode.

The RDcost calculation refers to the calculation of the rate-distortion value, that is, the image distortion value corresponding to a code rate.

In summary, the advantages of the present invention are:

1) Combining the characteristics of video texture and using the spatial correlation of video images, the prediction mode selection of the optimized intra-frame video encoder is optimized, which reduces the computational complexity of the prediction module and saves time while ensuring the prediction accuracy .

2) Combined with the recursive framework, the coding unit division is optimized to reduce unnecessary traversal and improve the coding speed.

3) According to the offline training experiments of different video sequences, the selection of the optimal number of prediction modes is determined, which ensures the stability of the coding efficiency of the algorithm for different video sequences.

The algorithm proposed by the present invention is tested on the AVS2 reference code RD_9.0, and adopts full intra-frame parameter configuration. The selection and parameters of the specific video sequence are in line with the requirements of the platform:

In order to evaluate the algorithm performance, the following three parameters are used for analysis:

PSNR(dB)＝PSNRY _pro -PSNRY _AVS2

BR, ET, and PSNR are respectively the increase percentage of the code rate, the percentage reduction of the encoding time, and the decrease value of the peak signal-to-noise ratio of the Y channel.

The experimental results are shown in Table 1 and Figure 2.

Table 1, algorithm experiment results of QP=43

As can be seen from Table 1, the algorithm proposed by the present invention reduces the encoding time by 48% to the existing latest AVS2 encoder algorithm, and has little impact on bit rate and signal-to-noise ratio, and the encoding quality of different video sequences is very stable.

In addition, the present invention is not limited to the above example applications, and can be implemented for all coding unit-based video coding methods. All references or transformations of the descriptions of the present invention under other coding standards shall belong to the protection scope of the appended claims of the present invention.

Claims (1)

1. a kind of optimization intraframe video coding method based on video texture information, it is characterised in that comprise the following steps：

The first step, all pixels to intracoded frame, i.e. I frames carry out gradient calculation, obtain representing the ladder in pixel motion direction Angle value；Gradient and predictive mode mapping relations figure are drawn by contrasting the relation of intra prediction direction figure and gradient, so as to compile The optimal prediction modes of each pixel are drawn during code；

Second step, attributive classification is carried out according to video texture information to encoding block, is then carried out with regard to the coding unit of same alike result Merge, then adjustable coding unit division is carried out to the coding unit after optimization；

3rd step, sets different mode candidate values according to the different attribute of coding unit and is predicted, then in gradient map The pattern for selecting respective amount from big to small according to the pattern frequency of occurrences in the gradient direction counted carries out RDcost calculating, The pattern for selecting RDcost minimum, as optimum prediction mode；

Described prediction, first by way of being trained under line, the different sequences of selection different resolution are tested respectively, root Coding rate and the optimal balance point of quality are selected according to coding result, then grouping comparison test, experiment is carried out around equalization point Select the optimal mode number of candidates that smooth unit and coarse cell should be selected respectively；

Described attributive classification refers to：The optimal prediction modes of each pixel of current coded unit are counted, when statistics knot There is a kind of model selection rate in fruit more than 80%, then judge in current coded unit, pixel has unified traffic direction, Image texture is relatively simple, is smooth unit by the attribute setup of the coding unit；Otherwise it is thick by the attribute setup of the unit Rough unit；

Described merging refers to：Analysis judgement is carried out from minimum coding unit 8x8 modules, when continuous four formed objects are compiled Code unit is smoothing properties and its main gradient direction has uniformity, then this four units can be combined into a larger list Member, and the large code unit is also set to smooth unit, gradient direction is consistent therewith；Otherwise selection current coded unit is big It is small, without union operation.