CN103905818B - Method for rapidly determining inter-frame prediction mode in HEVC standard based on Hough conversion - Google Patents

Abstract

The invention discloses a method for rapidly determining an inter-frame prediction coding mode in the latest national video compression standard, namely, the HEVC standard. The method includes the steps of firstly, conducting edge detection and Hough conversion on a current PU; secondly, conducting statistic analysis on the tangent value of the direction angle of a detected straight line so as to extract the most possible edge direction of the current PU according to the heuristic type information, optimally selecting candidate prediction modes so as to determine the MPM by conducting detection through the code rate and RMD process, and determining the final coding mode by conducting detection through the code rate and RDO process. Experiments show that the running speed of the HEVC can be increased by more than 20% according to the method.

Description

A fast determination method for intra prediction mode in HEVC standard based on Hough transform

technical field

The present invention relates to a method for speeding up the running speed of HEVC, the latest international compression standard in the field of video coding, and in particular to a method for speeding up the determination of the optimal coding mode of an intra-frame coding block by using technologies such as Hough transform Methods.

Background technique

HEVC (High Efficiency Video Coding) is the latest international video coding standard. It is jointly formulated by ISO/IEC Moving Picture Experts Group (MPEG) and ITU-T Video Coding Experts Group (VCEG). It is a high-performance video coding standard following H.264/AVC. Its encoding efficiency is more than 50% higher than that of H.264. HEVC still adopts the hybrid coding framework of prediction and transformation. Compared with the previous technology, the technology in this framework has not changed much, although the content and process of the general coding framework have not changed much, but in order to improve the video coding efficiency, the HEVC standard also Many new technologies have been added.

In the H.264 video coding standard, a macro block (Micro Block, MB) is the most basic unit of coding. However, in the HEVC coding standard, due to the popularity of high-definition video, it is difficult to meet the needs of high-definition video if such a fixed-size coding block is used for ultra-high-definition video reaching 2K and 4K. Therefore, HEVC does not continue to use the concept of MB, but uses a more flexible way to represent the block structure: coding unit (CU coding unit), prediction unit (PU prediction unit), and transform unit (TU transform unit). CU is the most basic division unit, its maximum coding size is 64×64, and its minimum coding size is 8×8. A PU is a basic unit that carries information related to the prediction process, and the PU size is smaller than or equal to the CU size. A CU may contain one or more prediction units PU of different sizes, and a PU may contain several transformation units TU, and a TU is a basic unit based on transformation and quantization. HEVC uses a quadtree structure to divide coding units.

In order to overcome the shortcomings of H.264/AVC, such as the small number of prediction modes, low prediction accuracy, and inaccuracy, HEVC greatly increases the number of available prediction modes. There are 35 prediction modes in HEVC intra prediction, among which mode 0 means to use planar (Planar) method for prediction, mode 1 means to use direct current (DC) method for prediction, and modes 2 to 34 represent to use various angles for prediction.

In the reference model HM4.0 of HEVC, the decision-making process of the intra mode is divided into three stages. First, rough selection of mode sets suitable for PUs of different sizes, calculate the rate-distortion cost (Rate-Distortion Cost, Rd-Cost) of all possible prediction modes, and arrange them in order of increasing cost, the one with the smallest Rd-Cost value The group prediction mode is the rough selection subset, and this process is called the rough selection mode determination (RMD) process. Here, when calculating the distortion, only the distortion of the prediction residual is considered, not the distortion after encoding and decoding. Second, it is verified whether the intra prediction modes of the left and upper reference blocks of the current block are included in the rough selection subset, and if not, they are added to the subset. Finally, rate-distortion optimization is performed on the prediction modes in the rough selection subset, and the mode with the least cost is the optimal prediction mode. This process is called the rate and distortion optimization (RDO) process. Here, in the RDO process, it is necessary to calculate the distortion of each candidate mode after encoding and decoding, so the calculation amount is very large. Since there are many optional intra-frame prediction modes in HEVC, the process of determining the intra-frame mode has a great computational complexity and is very time-consuming.

To sum up, with the increase in the number of optional prediction modes in HEVC, although the prediction accuracy and coding efficiency are improved, the coding complexity of HEVC is also greatly increased. Thus, the determination time of the intra-optimal prediction mode is greatly increased.

The present invention strives to reduce the time for determining the intra prediction mode in HEVC on the basis of maintaining the prediction accuracy. Since the edge information of the straight line in the image block can give us the trend direction of the pixels in the block, the present invention will use this heuristic information to simplify and further reduce the candidate intra prediction modes.

Contents of the invention

In view of the above-mentioned defects in the prior art, the present invention proposes a new method for determining fast intra-frame modes based on Hough transform. In order to achieve the purpose of reducing the amount of calculation while maintaining the accuracy of prediction as much as possible, the present invention will first perform edge detection on the image, then use Hough transform to search for straight line edges, and then determine the straight line edges in the current image prediction unit PU The most likely prediction direction of the current PU is determined by using the statistical histogram, and the most likely candidate prediction mode is determined according to this direction, so as to optimize RMD and code rate and distortion in the rough mode determination (RMD) process ( In the process of RDO), the candidate modes to be detected are reduced, the calculation amount is greatly reduced, and at the same time, a good prediction mode is selected as much as possible for the current PU. It is characterized in that, for each intra prediction unit PU in the video image, the method includes:

Step 1: For the current PU, determine whether its size is 4×4. If yes, perform mode sub-sampling with a step size of 1, and directly skip to step eight. Otherwise, use the Canny operator to operate on the current PU.

Step 2, according to the operation result of the Canny operator, the edge is thinned by the morphological method.

Step 3, select an appropriate threshold to extract and connect edge points in the PU.

Step 4: For the current PU, use the Hough transform to detect straight line edge pixel points, that is, keep the edge points on a certain straight line and remove the edge points on the curve.

Step five, for each straight line edge, determine its angle difference θ with the horizontal direction and its tangent value tanθ.

Step 6: According to the range of the tangent value of the prediction angle of each intra-frame prediction mode and the detected tangent value of each straight line edge in the current PU, use the statistical histogram to determine the straight line edge pixels falling within each range in the current PU sum of dots.

Step seven, according to the statistical histogram, determine several prediction modes with the largest sum of straight line edge pixels in the histogram.

Step 8, according to the size of the current PU and the obtained statistical histogram, determine the candidate intra-mode for entering the rough mode selection process RMD.

Step 9, according to the RMD process and the size of the current PU, determine the candidate intra-modes that enter the mode refinement process RDO.

Step ten, according to the result of RDO, determine the optimized intra prediction mode of the current PU.

Further, in the first step, mode sub-sampling is performed on the 4×4 block. In the HEVC reference model HM4.0, there are 19 suitable intra prediction modes for 4×4PU, and the directional prediction modes are: 7, 14, 6, 13, 1, 12, 5, 11, 4, 15, 8 , 16, 2, 17, 9, 18, 10. Mode subsampling on a 4×4 block retains only half of the modes, ie select modes 7, 6, 1, 5, 4, 8, 2, 9, 10.

The steps of Canny operator edge detection are as follows: Firstly, Gaussian filter is applied to smooth the image, and the mean value of the finite difference is calculated in the 3×3 neighborhood to calculate the magnitude and direction of the gray gradient.

Further, in the second step, the detected edge with a thick width is thinned using a morphological method, that is, a morphological corrosion operation is performed, and the eroded template element is

Further, in the third step, edge point extraction is performed by using threshold _{Th h} . An iterative method is used to select the optimal threshold. The implementation steps of the iterative method are as follows:

(1) Calculate the minimum value Z _min and maximum value Z _max in the current PU of the image, then the initial threshold T ₀ =(Z _min +Z _max )/2.

(2) According to the threshold T _k (k is the number of iterations), the image is divided into edge area and non-edge area, and the average value Z _O and Z _B of the two parts are calculated, namely

Here, h _i and h _j denote the values of pixels in the edge and non-edge regions, respectively. After calculating Z _O and Z _B , use the formula

T _k+1 ＝(Z _O +Z _B )/2

to calculate the new threshold T _k+1 .

(3) If T _k+1 =T _k , ie T _k is the desired threshold, then the algorithm ends here, otherwise go to step (2). The calculation is iterative until it converges to a stable threshold, which is the final result T _h .

Here, a dual-threshold method is employed to detect and connect the final edges from candidate edge points. When extracting edge points, the threshold used is T _h , and when connecting edge points, the threshold used should be smaller than T _h in order to connect edges. In the present invention, the threshold used in the connection of edge points is _Th /2. That is, after the first and second steps of processing, the pixel points whose value is greater than T _h are edge points. Then, using these edge points as seeds, search for adjacent pixel points with a value greater than _{Th h} /2, and add them as edge pixel points. Then, repeat this process until all the pixels in the PU are scanned once.

Table 1. Prediction angle, boundary angle range and tangent range of each prediction mode of 8×8, 16×16, 32×32 size PU

Table 2. Prediction angle, boundary angle range and tangent range of each prediction mode of 64×64 size PU

predictive model Forecast angle Boundary angle range Tangent range 1 -90 (-45, -135) (-∞, -1)U(1, +∞) 2 0 [-45, 45] [-1, 1]

Further, in step six, the statistical range of the tangent of the straight line edge angle is shown in Table 1 and Table 2. For example, as shown in Table 1, for candidate pattern 7, the statistical range of its angle tangent value is (-0.7417,-1.1033]. That is, the initial linear pixel point value in this range is set to 0, C ₇ =0, and then If the tangent value of the angle of the edge of a straight line falls within this range, then C ₇ =C ₇ +a _i , where a _i is the number of all pixels on this straight line in the current PU. This process is performed on all candidate modes , then the statistical histogram of each desired mode is obtained. Because mode 0 and mode 3 are non-directional prediction modes, they are not counted in table 1 and table 2, and they are all entered into the rough selection of RDO as candidate modes.

Table 3. The number of candidate prediction modes entering the RMD rough selection process in each PU size of HM4.0

4×4 19 8×8 35 16×16 35 32×32 35 64×64 4

Table 4. The number of candidate prediction modes entering the RMD rough selection process in each PU size of the present invention

4×4 9 8×8 7 16×16 5 32×32 3 64×64 1

Further, the process of Step 8 is shown in Table 4. For example, for a PU with a size of 16×16, 5 candidate modes with the largest value in the above statistical histogram will be selected. From the comparison of Table 3 and Table 4, it can be seen that the present invention greatly reduces the number of candidate patterns entering the RMD rough selection process. In this way, the calculation amount can be greatly reduced, and the operation speed of the HEVC standard can be accelerated.

Table 5. The number of candidate prediction modes entering the RDO selection process in each PU size of HM4.0

4×4 8 8×8 8 16×16 3 32×32 3 64×64 3

Table 6. The number of candidate prediction modes entering the RDO selection process in each PU size of the present invention

4×4 5 8×8 4 16×16 3 32×32 2 64×64 1

Further, the process of step nine is shown in Table 6. For example, for a PU with a size of 16×16, the best 3 candidate modes in the RMD process will be selected to enter the final RDO selection process. From the comparison of Table 5 and Table 6, it can be seen that the present invention greatly reduces the number of candidate patterns entering the RDO refinement process. In this way, the calculation amount can be greatly reduced, and the operation speed of the HEVC standard can be accelerated.

To sum up, the present invention proposes a new method for intra-frame prediction in HEVC, the latest international coding standard, so as to improve its running speed. The innovation of this method lies in: use Canny operator for edge detection, and use Hough transform for line edge detection, so as to extract the most likely edge direction of the current prediction unit according to these heuristic information. Then, according to these edge trends, the candidate prediction modes to be detected are selected and reduced. In this way, the computational complexity of HEVC encoding is greatly reduced and its running speed is improved.

The idea, specific structure and technical effects of the present invention will be further described below in conjunction with the accompanying drawings, so as to fully understand the purpose, features and effects of the present invention.

Description of drawings

FIG. 1 is a schematic diagram of an intra prediction mode in the HEVC standard;

Fig. 2 is the flow chart of the new method that the present invention proposes;

Fig. 3 is that the method proposed by the present invention and original method (HM4.0) compare the code rate and distortion performance curve of video test sequence Party Scene (832 * 480p resolution);

Fig. 4 is that the method proposed by the present invention and original method (HM4.0) are to the bit rate of video test sequence Slide Editing (720p resolution) and the comparative figure of distortion performance curve;

Fig. 5 is that the method proposed by the present invention and original method (HM4.0) compare the code rate and the degree of distortion performance curve of video test sequence Kimono1 (1080p resolution);

Fig. 6 is that the method proposed by the present invention and original method (HM4.0) compare the code rate and distortion performance curve of video test sequence SteamLocomotiveTrain (4k * 2k resolution);

detailed description

Below in conjunction with accompanying drawing, the embodiment of the present invention is described in detail: present embodiment implements under the premise of the technical scheme of the present invention, has provided detailed implementation and specific operation process, but protection scope of the present invention is not limited to the following the embodiment.

It can be seen from FIG. 1 that the original HEVC intra prediction has many candidate modes, and determining a more optimized mode has relatively large computational complexity.

The present invention utilizes the C++ language on the basis of the HM4.0 program of HEVC, and utilizes the OpenCV function library to realize and experiment the proposed method.

As shown in accompanying drawing 2, the method for reducing the computational complexity of its intra prediction of the present invention comprises the following steps:

Step 1: For the current PU, determine whether its size is 4×4. If yes, perform mode sub-sampling with a step size of 1, and directly skip to step eight. Otherwise, use the Canny operator to operate on the current PU. For a 4×4 PU, because its size is too small, the correlation between the direction of the straight line edge extracted from it and the direction of the prediction mode is not particularly large, so skip directly to step 8. For a PU whose size is larger than 4×4, the correlation between the direction of the straight line edge extracted from it and the direction of the prediction mode is very large, so it can be used to determine the candidate mode to be detected. In OpenCV, the function cvCanny is used to access the Canny operator for edge detection. The function call prototype is void cvCanny(const CvArr* image, CvArr*edges, double threshold1, double threshold2, int aperture_size=3). The function cvCanny uses the Canny algorithm to search for edges in the input image and outputs these edges identified in the image.

Step 2, according to the operation result of the Canny operator, the edge is thinned by the morphological method. ready to use as follows

The template element performs morphological erosion on edges with width greater than 3.

Step 3, select an appropriate threshold to extract and connect edge points in the PU. Here, threshold _Th is used to extract edge points, and an iterative method is used to select the best threshold. The implementation steps of the iterative method are as follows:

Calculate the minimum value Z _min and maximum value Z _max in the current PU of the image, then the initial threshold T ₀ =(Z _min +Z _max )/2.

According to the threshold T _k (k is the number of iterations), the image is divided into edge area and non-edge area, and the average value Z _O and Z _B of the two parts are calculated, namely

Here, h _i and h _j denote the values of pixels in the edge and non-edge regions, respectively. After calculating Z _O and Z _B , use the formula

T _k+1 ＝(Z _O +Z _B )/2

to calculate the new threshold T _K+1 .

If T _k+1 =T _K , that is, T _k is the desired threshold, then the algorithm ends here, otherwise go to step (2). The calculation is iterative until it converges to a stable threshold, which is the final result T _h .

Here, a dual-threshold method is employed to detect and connect the final edges from candidate edge points. When extracting edge points, the threshold used is T _h , and when connecting edge points, the threshold used should be smaller than T _h in order to connect edges. In the present invention, the threshold used in the connection of edge points is _Th /2. That is, after the first and second steps of processing, the pixel points whose value is greater than T _h are edge points. Then, using these edge points as seeds, search for adjacent pixel points with a value greater than _{Th h} /2, and add them as edge pixel points. Then, repeat this process until all the pixels in the PU are scanned once.

Step 4: For the current PU, use the Hough transform to detect straight line edge pixel points, that is, keep the edge points on a certain straight line and remove the edge points on the curve. In OpenCV, the function cvHoughLines2 is used to detect straight line segments. The function call prototype of cvHoughLines2 is CvSeq* cvHoughLines2(CvArr* image, void* linestorage, int method, double rho, double theta, int threshold, double param1, doubleparam2). In the present invention, the parameter method is set to CV_HOUGH_PROBABILISTIC to indicate that the probability Hough transform is selected to detect the straight line segment. threshold is the threshold parameter, if the corresponding cumulative value is greater than it, the function will return this line segment. param1 indicates the minimum line segment length, and param2 indicates the maximum interval value for connecting line segments on the same straight line.

Step five, for each straight line edge, determine its angle difference θ with the horizontal direction and its tangent value tanθ.

Step 6: According to the range of the tangent value of the prediction angle of each intra-frame prediction mode and the detected tangent value of each straight line edge in the current PU, use the statistical histogram to determine the straight line edge pixels falling within each range in the current PU sum of dots. Table 1 and Table 2 show the statistical range of the tangent value of the straight line edge angle. For example, as shown in Table 1, for candidate pattern 7, the statistical range of its angle tangent value is (-0.7417,-1.1033]. That is, the initial linear pixel point value in this range is set to 0, C ₇ =0, and then If the tangent value of the angle of a straight line edge falls within this range (-0.7417, -1.1033], then C ₇ =C ₇ +a _i , where a _i is the number of all pixels on this straight line in the current PU. Right All candidate modes are all carried out this processing, then obtained the statistical histogram of each mode that wants.Because mode 0, mode 3 are non-directional prediction modes, so are not counted in table 1, table 2, they are all fixed as candidate Mode entry into RDO's rough selection.

Step seven, according to the statistical histogram, determine several prediction modes with the largest sum of straight line edge pixels in the histogram.

Step 8, according to the size of the current PU and the obtained statistical histogram, determine the candidate intra-mode for entering the rough mode selection process RMD. Further, for a PU of a certain size, the number of candidate prediction modes selected for RDO detection is shown in Table 6. For example, for a PU with a size of 16×16, 5 candidate modes with the largest value in the above statistical histogram will be selected. From the comparison of Table 3 and Table 4, it can be seen that the present invention greatly reduces the number of candidate patterns entering the RMD rough selection process. In this way, the calculation amount can be greatly reduced, and the operation speed of the HEVC standard can be accelerated.

Step 9, according to the RMD process and the size of the current PU, determine the candidate intra-modes that enter the mode refinement process RDO. Further, for a PU of a certain size, the number of candidate prediction modes selected for RDO detection is shown in Table 6. For example, for a PU with a size of 16×16, the best 3 candidate modes in the RMD process will be selected to enter the final RDO selection process. From the comparison of Table 5 and Table 6, it can be seen that the present invention greatly reduces the number of candidate patterns entering the RDO rough selection process. In this way, the calculation amount can be greatly reduced, and the operation speed of the HEVC standard can be accelerated.

Step ten, according to the result of RDO, determine the optimized intra prediction mode of the current PU.

The present invention tested the proposed method according to the common test environment. The encoding format is all I-frame structure, the test QP points are 37, 32, 27, and 22, and the video sequence is entropy encoded using CABAC (Context-based Adaptive Binary Arithmetic Coding). Considering that HEVC will be widely used in high-resolution video sequences in the future, the test is mainly based on high-definition and standard-definition test sequences, mainly including 25601×600p, 1920×1080p, 1280×720p and 832×480p, a total of 4 video sizes test sequence. Here we mainly consider the performance of this method from the reduction of coding time brought by the proposed new method and the corresponding cost. The performance index mainly expresses the performance of the proposed method by parameters such as BD-PSNR/Rate, ΔB, ΔP, and ΔT.

Table 7. Experimental test results of the proposed method of the present invention

Experimental result is as shown in table 7, if it is a positive number in all data results, it means that the method of the present invention under this index is increased relative to its numerical value of the original method, if it is a negative number, it means that under this index, the method of the present invention Compared with the original method, its value is reduced. The metrics for testing performance are defined as follows:

(1) The saved encoding time ΔT, namely

Among them, T _HM4.0 represents the encoding time of using the existing algorithm of HM4.0, and T _Proposed is the encoding time of the algorithm proposed in this paper. Under the condition of obtaining the same coding efficiency, the larger ΔT means that the computational complexity of the coding end is reduced more, and the performance of the proposed method is better.

(2) The code rate difference ΔB between the proposed method and the HM4.0 intra prediction algorithm, namely

Wherein, B _HM4.0 represents the coding rate of the existing HM4.0 method, and B _Proposed is the coding rate of the method proposed by the present invention. The smaller the ΔB, the closer the coding performance of the proposed method to the existing method of HM4.0.

(3) The difference ΔP between the proposed method and the peak signal-to-noise ratio (Peak Signal to Noise Ratio, PSNR) of the luminance component of the HM4.0 intra prediction method, namely

ΔP＝P _Proposed -P _HM4.0

Wherein, P _HM4.0 represents the peak signal-to-noise ratio PSNR of the luminance component of the existing method of HM4.0, and P _Proposed is the PSNR of the luminance component of the method proposed by the present invention. The smaller the ΔP, the closer the coding performance of the proposed method to the existing method of HM4.0.

BD-Rate indicates the increase value of the code rate under the same peak signal-to-noise ratio (PSNR). BD-PSNR indicates the reduction value of PSNR under the same code rate.

It can be seen from the experimental results in Table 7 that the proposed method has only a very small decrease in peak signal-to-noise ratio (PSNR) and a very small increase in code rate, so it has better performance. At the same time, it can be seen from the table that the method proposed by the present invention can save the encoding time by a minimum of 16.15% and a maximum of 32.73% compared with the standard method in HM4.0.

The curves of bit rate and distortion (Rate-Distortion) of the proposed method and the original method (HM4.0) for video test sequences at different resolutions obtained from the experimental results are shown in Figures 3, 4, 5, and 6 . The abscissa of these graphs is the code rate, and the unit is the number of thousand bits to be transmitted per second (kbps). The ordinate of these graphs is the peak signal-to-noise ratio (PSNR) in dB. From these figures, it can be seen that the two curves of code rate and distortion degree of the method proposed by the present invention and the original method are basically the same. From this, it can be explained that the proposed method maintains the accuracy of the prediction mode, and under the same code rate condition, the degree of distortion is not deteriorated or is relatively greatly reduced.

From Table 7 and Figures 3, 4, 5, and 6, it can be seen that the proposed method can save an average of 23% of the encoding time compared to the original method. At the same time, the fluctuation of its code rate and PSNR is small, and the reduction in performance is almost negligible compared to the computational cost saved. It can be seen that the method proposed by the present invention has better performance.

The preferred specific embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make many modifications and changes according to the concept of the present invention without creative efforts. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning or limited experiments on the basis of the prior art shall be within the scope of protection defined by the claims.

Claims (7)

1. A method for quickly determining an intra-frame prediction coding mode in the international video compression standard HEVC, the method comprising: Step 1. For the current prediction unit (PU), judge whether its size is 4×4. If so, perform mode sub-sampling with a step size of 1, and directly skip to step 8; otherwise, use the Canny operator to perform operation; Step 2, according to the result of the operation of the Canny operator, refine the edge with a morphological method; Step 3, select an appropriate threshold to extract and connect edge points in the PU; Step 4, for the current PU, use Hough transform to detect straight line edge pixel points, that is, keep the edge points on a certain straight line, and remove the edge points on the curve; Step five, for each straight line edge, determine its angle difference θ with the horizontal direction, and its tangent value tanθ; Step 6: According to the range of the tangent value of the prediction angle of each intra-frame prediction mode and the detected tangent value of each straight line edge in the current PU, use the statistical histogram to determine the straight line edge pixels falling within each range in the current PU sum of points; Step 7, according to the statistical histogram, determine several prediction modes with the largest sum of straight line edge pixels in the histogram; Step 8, according to the size of the current PU and the obtained statistical histogram, determine the candidate intra-mode for entering the mode rough selection process (RMD); Step 9, according to the RMD process and the size of the current PU, determine the candidate intra-mode for entering the mode selection process (RDO); Step ten, according to the result of RDO, determine the optimized intra prediction mode of the current PU. 2. The method as claimed in claim 1, adopting a morphological method in said step 2 to refine the detected width greater than 3 edges, that is, to carry out morphological corrosion operations, the template elements of which are corroded are $\left[\begin{array}{ccc}\mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\end{array}\right]<span\; class="notranslate">\; .</span>$ 3. method as claimed in claim 1, adopt threshold value _Th to carry out edge point extraction in described step 3, adopt iterative method to choose optimal threshold value, iterative method realization step is as follows: (1) Find the minimum value Z _min and the maximum value Z _max in the current PU of the image, then the initial threshold T ₀ = (Z _min + Z _max) /2, (2) According to the threshold T _k (k is the number of iterations), the image is divided into edge pixels and non-edge pixels, and the average value Z _O and Z _B of the two parts are calculated, namely ${\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; o</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; \&CenterDot;</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; /</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}$ ${\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; B</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 255</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; \&CenterDot;</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; /</span>\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 255</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}$ Here, h _i and h _j represent the values of edge pixels and non-edge pixels respectively. After calculating Z _O and Z _B , use the formula T _k+1 ＝(Z _O +Z _B )/2 to calculate the new threshold T _k+1 ; (3) If T _k+1 = T _k , that is, T _k is the desired threshold, the algorithm ends here; otherwise, go to step (2) and iteratively calculate until it converges to a stable threshold, which is the final Result T _h . 4. The method according to claim 1, when extracting edge points in said step 3, the threshold used is _Th , and the threshold used when connecting edge points should be smaller than _Th to connect the edges; The threshold used in the connection of edge points in the present invention is T _h /2, that is, after the first and second steps of processing, the pixel points with a value greater than T _h are edge points; then use these edge points as seeds to find The adjacent pixels whose value is greater than T _h /2 are added as edge pixels; then, this process is repeated until all the pixels in the PU are scanned once. 5. The method according to claim 1, in said step 6, according to the range of the tangent of the prediction angle of each intra-frame prediction mode, and the detected tangent of each straight line edge in the current PU, use a statistical histogram Determine the sum of straight-line edge pixels falling within each range in the current PU to determine the edge direction of the current PU. 6. The method as claimed in claim 1, in said step eight, to sizes 4×4, 8×8, 16×16, 32×32, and 64×64, respectively select 9, 7, 5, 3, One candidate mode enters the RMD detection process; among them, for prediction units with PU sizes of 8×8, 16×16, 32×32, and 64×64, the candidate with the largest sum of straight-line edge pixels in the statistical histogram is selected Patterns are entered into the RMD detection process to determine the patterns that enter the RMD detection process. 7. The method according to claim 1, in said step 9, for sizes 4×4, 8×8, 16×16, 32×32, and 64×64, respectively select 5, 4, 3, 2, One candidate mode enters the RDO detection process; after the selection standard is the RMD process, the candidate mode with a small code rate and prediction distortion enters the RDO detection process to determine the candidate mode that enters the RDO detection process.